[clang] [llvm] [x86][AVX-VNNI] Fix VPDPWXXD Argument Types (PR #169456)

Simon Pilgrim via llvm-commits llvm-commits at lists.llvm.org
Tue Dec 9 08:34:31 PST 2025


https://github.com/RKSimon updated https://github.com/llvm/llvm-project/pull/169456

>From 3735e20c2c20a426ae14c196f08468c41a347694 Mon Sep 17 00:00:00 2001
From: BaiXilin <x53bai at uwaterloo.ca>
Date: Tue, 25 Nov 2025 00:18:36 -0500
Subject: [PATCH 1/4] [x86][AVX-VNNI] Fix VPDPWXXD Argument Types

Fixed the argument types of the following intrinsics to match with the
ISA:
 - vpdpwssd_128, vpdpwssd_256, vpdpwssd_512,
 - vpdpwssds_128, vpdpwssds_256, vpdpwssds_512
 - more to come
---
 clang/include/clang/Basic/BuiltinsX86.td      | 12 ++--
 clang/lib/Headers/avx512vlvnniintrin.h        | 17 ++---
 clang/lib/Headers/avx512vnniintrin.h          |  8 +--
 clang/lib/Headers/avxvnniintrin.h             | 12 ++--
 .../test/CodeGen/X86/avx512vlvnni-builtins.c  | 24 +++----
 clang/test/CodeGen/X86/avx512vnni-builtins.c  | 12 ++--
 clang/test/CodeGen/X86/avxvnni-builtins.c     | 16 ++---
 llvm/include/llvm/IR/IntrinsicsX86.td         | 24 +++----
 llvm/lib/IR/AutoUpgrade.cpp                   | 72 +++++++++++++++++++
 .../Instrumentation/MemorySanitizer.cpp       | 36 +++++-----
 .../CodeGen/X86/avx512vl_vnni-intrinsics.ll   | 58 +++++++--------
 .../X86/avx512vnni-intrinsics-upgrade.ll      | 42 ++++++++---
 .../test/CodeGen/X86/avx512vnni-intrinsics.ll | 28 ++++----
 .../X86/avx_vnni-intrinsics-upgrade.ll        | 44 ++++++++++++
 llvm/test/CodeGen/X86/avx_vnni-intrinsics.ll  | 24 +++----
 .../CodeGen/X86/stack-folding-int-avxvnni.ll  | 40 +++++------
 .../X86/avx512vl_vnni-intrinsics-upgrade.ll   | 72 +++++++++----------
 .../X86/avx512vl_vnni-intrinsics.ll           | 72 +++++++++----------
 .../X86/avx512vnni-intrinsics-upgrade.ll      | 36 +++++-----
 .../X86/avx512vnni-intrinsics.ll              | 36 +++++-----
 .../X86/avx_vnni-intrinsics.ll                | 24 +++----
 21 files changed, 426 insertions(+), 283 deletions(-)

diff --git a/clang/include/clang/Basic/BuiltinsX86.td b/clang/include/clang/Basic/BuiltinsX86.td
index cb08e2107f072..f61dd0546cc24 100644
--- a/clang/include/clang/Basic/BuiltinsX86.td
+++ b/clang/include/clang/Basic/BuiltinsX86.td
@@ -1132,27 +1132,27 @@ let Features = "avx512vnni", Attributes = [NoThrow, Const, RequiredVectorWidth<5
 }
 
 let Features = "avx512vl,avx512vnni|avxvnni", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in {
-  def vpdpwssd128 : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<4, int>, _Vector<4, int>)">;
+  def vpdpwssd128 : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<8, short>, _Vector<8, short>)">;
 }
 
 let Features = "avx512vl,avx512vnni|avxvnni", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in {
-  def vpdpwssd256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<8, int>, _Vector<8, int>)">;
+  def vpdpwssd256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<16, short>, _Vector<16, short>)">;
 }
 
 let Features = "avx512vnni", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in {
-  def vpdpwssd512 : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<16, int>, _Vector<16, int>)">;
+  def vpdpwssd512 : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<32, short>, _Vector<32, short>)">;
 }
 
 let Features = "avx512vl,avx512vnni|avxvnni", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in {
-  def vpdpwssds128 : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<4, int>, _Vector<4, int>)">;
+  def vpdpwssds128 : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<8, short>, _Vector<8, short>)">;
 }
 
 let Features = "avx512vl,avx512vnni|avxvnni", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in {
-  def vpdpwssds256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<8, int>, _Vector<8, int>)">;
+  def vpdpwssds256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<16, short>, _Vector<16, short>)">;
 }
 
 let Features = "avx512vnni", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in {
-  def vpdpwssds512 : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<16, int>, _Vector<16, int>)">;
+  def vpdpwssds512 : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<32, short>, _Vector<32, short>)">;
 }
 
 let Features = "avxvnniint8|avx10.2", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in {
diff --git a/clang/lib/Headers/avx512vlvnniintrin.h b/clang/lib/Headers/avx512vlvnniintrin.h
index a1a0338a69e0d..4b8a199af32e5 100644
--- a/clang/lib/Headers/avx512vlvnniintrin.h
+++ b/clang/lib/Headers/avx512vlvnniintrin.h
@@ -80,8 +80,8 @@
 ///    ENDFOR
 ///    DST[MAX:256] := 0
 /// \endcode
-#define _mm256_dpwssd_epi32(S, A, B) \
-  ((__m256i)__builtin_ia32_vpdpwssd256((__v8si)(S), (__v8si)(A), (__v8si)(B)))
+#define _mm256_dpwssd_epi32(S, A, B)                                           \
+  ((__m256i)__builtin_ia32_vpdpwssd256((__v8si)(S), (__v16hi)(A), (__v16hi)(B)))
 
 /// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a A with
 /// corresponding 16-bit integers in \a B, producing 2 intermediate signed 32-bit
@@ -98,8 +98,9 @@
 ///    ENDFOR
 ///    DST[MAX:256] := 0
 /// \endcode
-#define _mm256_dpwssds_epi32(S, A, B) \
-  ((__m256i)__builtin_ia32_vpdpwssds256((__v8si)(S), (__v8si)(A), (__v8si)(B)))
+#define _mm256_dpwssds_epi32(S, A, B)                                          \
+  ((__m256i)__builtin_ia32_vpdpwssds256((__v8si)(S), (__v16hi)(A),             \
+                                        (__v16hi)(B)))
 
 /// Multiply groups of 4 adjacent pairs of unsigned 8-bit integers in \a A with
 /// corresponding signed 8-bit integers in \a B, producing 4 intermediate signed
@@ -157,8 +158,8 @@
 ///    ENDFOR
 ///    DST[MAX:128] := 0
 /// \endcode
-#define _mm_dpwssd_epi32(S, A, B) \
-  ((__m128i)__builtin_ia32_vpdpwssd128((__v4si)(S), (__v4si)(A), (__v4si)(B)))
+#define _mm_dpwssd_epi32(S, A, B)                                              \
+  ((__m128i)__builtin_ia32_vpdpwssd128((__v4si)(S), (__v8hi)(A), (__v8hi)(B)))
 
 /// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a A with
 /// corresponding 16-bit integers in \a B, producing 2 intermediate signed 32-bit
@@ -175,8 +176,8 @@
 ///    ENDFOR
 ///    DST[MAX:128] := 0
 /// \endcode
-#define _mm_dpwssds_epi32(S, A, B) \
-  ((__m128i)__builtin_ia32_vpdpwssds128((__v4si)(S), (__v4si)(A), (__v4si)(B)))
+#define _mm_dpwssds_epi32(S, A, B)                                             \
+  ((__m128i)__builtin_ia32_vpdpwssds128((__v4si)(S), (__v8hi)(A), (__v8hi)(B)))
 
 static __inline__ __m256i __DEFAULT_FN_ATTRS256
 _mm256_mask_dpbusd_epi32(__m256i __S, __mmask8 __U, __m256i __A, __m256i __B)
diff --git a/clang/lib/Headers/avx512vnniintrin.h b/clang/lib/Headers/avx512vnniintrin.h
index c386923360de6..2ce88efe4a04f 100644
--- a/clang/lib/Headers/avx512vnniintrin.h
+++ b/clang/lib/Headers/avx512vnniintrin.h
@@ -68,8 +68,8 @@ _mm512_maskz_dpbusds_epi32(__mmask16 __U, __m512i __S, __m512i __A, __m512i __B)
 static __inline__ __m512i __DEFAULT_FN_ATTRS
 _mm512_dpwssd_epi32(__m512i __S, __m512i __A, __m512i __B)
 {
-  return (__m512i)__builtin_ia32_vpdpwssd512((__v16si)__S, (__v16si)__A,
-                                             (__v16si)__B);
+  return (__m512i)__builtin_ia32_vpdpwssd512((__v16si)__S, (__v32hi)__A,
+                                             (__v32hi)__B);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
@@ -91,8 +91,8 @@ _mm512_maskz_dpwssd_epi32(__mmask16 __U, __m512i __S, __m512i __A, __m512i __B)
 static __inline__ __m512i __DEFAULT_FN_ATTRS
 _mm512_dpwssds_epi32(__m512i __S, __m512i __A, __m512i __B)
 {
-  return (__m512i)__builtin_ia32_vpdpwssds512((__v16si)__S, (__v16si)__A,
-                                              (__v16si)__B);
+  return (__m512i)__builtin_ia32_vpdpwssds512((__v16si)__S, (__v32hi)__A,
+                                              (__v32hi)__B);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
diff --git a/clang/lib/Headers/avxvnniintrin.h b/clang/lib/Headers/avxvnniintrin.h
index 3c4c44a930fe2..1d2e8c906effc 100644
--- a/clang/lib/Headers/avxvnniintrin.h
+++ b/clang/lib/Headers/avxvnniintrin.h
@@ -109,7 +109,8 @@ _mm256_dpbusds_avx_epi32(__m256i __S, __m256i __A, __m256i __B)
 static __inline__ __m256i __DEFAULT_FN_ATTRS256
 _mm256_dpwssd_avx_epi32(__m256i __S, __m256i __A, __m256i __B)
 {
-  return (__m256i)__builtin_ia32_vpdpwssd256((__v8si)__S, (__v8si)__A, (__v8si)__B);
+  return (__m256i)__builtin_ia32_vpdpwssd256((__v8si)__S, (__v16hi)__A,
+                                             (__v16hi)__B);
 }
 
 /// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a __A with
@@ -130,7 +131,8 @@ _mm256_dpwssd_avx_epi32(__m256i __S, __m256i __A, __m256i __B)
 static __inline__ __m256i __DEFAULT_FN_ATTRS256
 _mm256_dpwssds_avx_epi32(__m256i __S, __m256i __A, __m256i __B)
 {
-  return (__m256i)__builtin_ia32_vpdpwssds256((__v8si)__S, (__v8si)__A, (__v8si)__B);
+  return (__m256i)__builtin_ia32_vpdpwssds256((__v8si)__S, (__v16hi)__A,
+                                              (__v16hi)__B);
 }
 
 /// Multiply groups of 4 adjacent pairs of unsigned 8-bit integers in \a __A with
@@ -199,7 +201,8 @@ _mm_dpbusds_avx_epi32(__m128i __S, __m128i __A, __m128i __B)
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_dpwssd_avx_epi32(__m128i __S, __m128i __A, __m128i __B)
 {
-  return (__m128i)__builtin_ia32_vpdpwssd128((__v4si)__S, (__v4si)__A, (__v4si)__B);
+  return (__m128i)__builtin_ia32_vpdpwssd128((__v4si)__S, (__v8hi)__A,
+                                             (__v8hi)__B);
 }
 
 /// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a __A with
@@ -220,7 +223,8 @@ _mm_dpwssd_avx_epi32(__m128i __S, __m128i __A, __m128i __B)
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_dpwssds_avx_epi32(__m128i __S, __m128i __A, __m128i __B)
 {
-  return (__m128i)__builtin_ia32_vpdpwssds128((__v4si)__S, (__v4si)__A, (__v4si)__B);
+  return (__m128i)__builtin_ia32_vpdpwssds128((__v4si)__S, (__v8hi)__A,
+                                              (__v8hi)__B);
 }
 
 #undef __DEFAULT_FN_ATTRS128
diff --git a/clang/test/CodeGen/X86/avx512vlvnni-builtins.c b/clang/test/CodeGen/X86/avx512vlvnni-builtins.c
index f63b5c6e73917..11dbd717a9f77 100644
--- a/clang/test/CodeGen/X86/avx512vlvnni-builtins.c
+++ b/clang/test/CodeGen/X86/avx512vlvnni-builtins.c
@@ -47,41 +47,41 @@ __m256i test_mm256_dpbusds_epi32(__m256i __S, __m256i __A, __m256i __B) {
 
 __m256i test_mm256_mask_dpwssd_epi32(__m256i __S, __mmask8 __U, __m256i __A, __m256i __B) {
   // CHECK-LABEL: test_mm256_mask_dpwssd_epi32
-  // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+  // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
   // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_mask_dpwssd_epi32(__S, __U, __A, __B);
 }
 
 __m256i test_mm256_maskz_dpwssd_epi32(__mmask8 __U, __m256i __S, __m256i __A, __m256i __B) {
   // CHECK-LABEL: test_mm256_maskz_dpwssd_epi32
-  // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+  // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
   // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_maskz_dpwssd_epi32(__U, __S, __A, __B);
 }
 
 __m256i test_mm256_dpwssd_epi32(__m256i __S, __m256i __A, __m256i __B) {
   // CHECK-LABEL: test_mm256_dpwssd_epi32
-  // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+  // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
   return _mm256_dpwssd_epi32(__S, __A, __B);
 }
 
 __m256i test_mm256_mask_dpwssds_epi32(__m256i __S, __mmask8 __U, __m256i __A, __m256i __B) {
   // CHECK-LABEL: test_mm256_mask_dpwssds_epi32
-  // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+  // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
   // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_mask_dpwssds_epi32(__S, __U, __A, __B);
 }
 
 __m256i test_mm256_maskz_dpwssds_epi32(__mmask8 __U, __m256i __S, __m256i __A, __m256i __B) {
   // CHECK-LABEL: test_mm256_maskz_dpwssds_epi32
-  // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+  // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
   // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_maskz_dpwssds_epi32(__U, __S, __A, __B);
 }
 
 __m256i test_mm256_dpwssds_epi32(__m256i __S, __m256i __A, __m256i __B) {
   // CHECK-LABEL: test_mm256_dpwssds_epi32
-  // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+  // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
   return _mm256_dpwssds_epi32(__S, __A, __B);
 }
 
@@ -127,41 +127,41 @@ __m128i test_mm_dpbusds_epi32(__m128i __S, __m128i __A, __m128i __B) {
 
 __m128i test_mm_mask_dpwssd_epi32(__m128i __S, __mmask8 __U, __m128i __A, __m128i __B) {
   // CHECK-LABEL: test_mm_mask_dpwssd_epi32
-  // CHECK: call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+  // CHECK: call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
   // CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
   return _mm_mask_dpwssd_epi32(__S, __U, __A, __B);
 }
 
 __m128i test_mm_maskz_dpwssd_epi32(__mmask8 __U, __m128i __S, __m128i __A, __m128i __B) {
   // CHECK-LABEL: test_mm_maskz_dpwssd_epi32
-  // CHECK: call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+  // CHECK: call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
   // CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
   return _mm_maskz_dpwssd_epi32(__U, __S, __A, __B);
 }
 
 __m128i test_mm_dpwssd_epi32(__m128i __S, __m128i __A, __m128i __B) {
   // CHECK-LABEL: test_mm_dpwssd_epi32
-  // CHECK: call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+  // CHECK: call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
   return _mm_dpwssd_epi32(__S, __A, __B);
 }
 
 __m128i test_mm_mask_dpwssds_epi32(__m128i __S, __mmask8 __U, __m128i __A, __m128i __B) {
   // CHECK-LABEL: test_mm_mask_dpwssds_epi32
-  // CHECK: call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+  // CHECK: call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
   // CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
   return _mm_mask_dpwssds_epi32(__S, __U, __A, __B);
 }
 
 __m128i test_mm_maskz_dpwssds_epi32(__mmask8 __U, __m128i __S, __m128i __A, __m128i __B) {
   // CHECK-LABEL: test_mm_maskz_dpwssds_epi32
-  // CHECK: call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+  // CHECK: call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
   // CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
   return _mm_maskz_dpwssds_epi32(__U, __S, __A, __B);
 }
 
 __m128i test_mm_dpwssds_epi32(__m128i __S, __m128i __A, __m128i __B) {
   // CHECK-LABEL: test_mm_dpwssds_epi32
-  // CHECK: call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+  // CHECK: call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
   return _mm_dpwssds_epi32(__S, __A, __B);
 }
 
diff --git a/clang/test/CodeGen/X86/avx512vnni-builtins.c b/clang/test/CodeGen/X86/avx512vnni-builtins.c
index afe80458e37cc..6b8465206eedb 100644
--- a/clang/test/CodeGen/X86/avx512vnni-builtins.c
+++ b/clang/test/CodeGen/X86/avx512vnni-builtins.c
@@ -47,41 +47,41 @@ __m512i test_mm512_dpbusds_epi32(__m512i __S, __m512i __A, __m512i __B) {
 
 __m512i test_mm512_mask_dpwssd_epi32(__m512i __S, __mmask16 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: test_mm512_mask_dpwssd_epi32
-  // CHECK: call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+  // CHECK: call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
   // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
   return _mm512_mask_dpwssd_epi32(__S, __U, __A, __B);
 }
 
 __m512i test_mm512_maskz_dpwssd_epi32(__mmask16 __U, __m512i __S, __m512i __A, __m512i __B) {
   // CHECK-LABEL: test_mm512_maskz_dpwssd_epi32
-  // CHECK: call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+  // CHECK: call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
   // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
   return _mm512_maskz_dpwssd_epi32(__U, __S, __A, __B);
 }
 
 __m512i test_mm512_dpwssd_epi32(__m512i __S, __m512i __A, __m512i __B) {
   // CHECK-LABEL: test_mm512_dpwssd_epi32
-  // CHECK: call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+  // CHECK: call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
   return _mm512_dpwssd_epi32(__S, __A, __B);
 }
 
 __m512i test_mm512_mask_dpwssds_epi32(__m512i __S, __mmask16 __U, __m512i __A, __m512i __B) {
   // CHECK-LABEL: test_mm512_mask_dpwssds_epi32
-  // CHECK: call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+  // CHECK: call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
   // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
   return _mm512_mask_dpwssds_epi32(__S, __U, __A, __B);
 }
 
 __m512i test_mm512_maskz_dpwssds_epi32(__mmask16 __U, __m512i __S, __m512i __A, __m512i __B) {
   // CHECK-LABEL: test_mm512_maskz_dpwssds_epi32
-  // CHECK: call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+  // CHECK: call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
   // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
   return _mm512_maskz_dpwssds_epi32(__U, __S, __A, __B);
 }
 
 __m512i test_mm512_dpwssds_epi32(__m512i __S, __m512i __A, __m512i __B) {
   // CHECK-LABEL: test_mm512_dpwssds_epi32
-  // CHECK: call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+  // CHECK: call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
   return _mm512_dpwssds_epi32(__S, __A, __B);
 }
 
diff --git a/clang/test/CodeGen/X86/avxvnni-builtins.c b/clang/test/CodeGen/X86/avxvnni-builtins.c
index 7948e0d57d9bf..6557a26807eb2 100644
--- a/clang/test/CodeGen/X86/avxvnni-builtins.c
+++ b/clang/test/CodeGen/X86/avxvnni-builtins.c
@@ -19,13 +19,13 @@ __m256i test_mm256_dpbusds_epi32(__m256i __S, __m256i __A, __m256i __B) {
 
 __m256i test_mm256_dpwssd_epi32(__m256i __S, __m256i __A, __m256i __B) {
   // CHECK-LABEL: test_mm256_dpwssd_epi32
-  // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+  // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
   return _mm256_dpwssd_epi32(__S, __A, __B);
 }
 
 __m256i test_mm256_dpwssds_epi32(__m256i __S, __m256i __A, __m256i __B) {
   // CHECK-LABEL: test_mm256_dpwssds_epi32
-  // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+  // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
   return _mm256_dpwssds_epi32(__S, __A, __B);
 }
 
@@ -43,13 +43,13 @@ __m128i test_mm_dpbusds_epi32(__m128i __S, __m128i __A, __m128i __B) {
 
 __m128i test_mm_dpwssd_epi32(__m128i __S, __m128i __A, __m128i __B) {
   // CHECK-LABEL: test_mm_dpwssd_epi32
-  // CHECK: call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+  // CHECK: call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
   return _mm_dpwssd_epi32(__S, __A, __B);
 }
 
 __m128i test_mm_dpwssds_epi32(__m128i __S, __m128i __A, __m128i __B) {
   // CHECK-LABEL: test_mm_dpwssds_epi32
-  // CHECK: call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+  // CHECK: call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
   return _mm_dpwssds_epi32(__S, __A, __B);
 }
 
@@ -67,13 +67,13 @@ __m256i test_mm256_dpbusds_avx_epi32(__m256i __S, __m256i __A, __m256i __B) {
 
 __m256i test_mm256_dpwssd_avx_epi32(__m256i __S, __m256i __A, __m256i __B) {
   // CHECK-LABEL: test_mm256_dpwssd_avx_epi32
-  // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+  // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
   return _mm256_dpwssd_avx_epi32(__S, __A, __B);
 }
 
 __m256i test_mm256_dpwssds_avx_epi32(__m256i __S, __m256i __A, __m256i __B) {
   // CHECK-LABEL: test_mm256_dpwssds_avx_epi32
-  // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+  // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
   return _mm256_dpwssds_avx_epi32(__S, __A, __B);
 }
 
@@ -91,12 +91,12 @@ __m128i test_mm_dpbusds_avx_epi32(__m128i __S, __m128i __A, __m128i __B) {
 
 __m128i test_mm_dpwssd_avx_epi32(__m128i __S, __m128i __A, __m128i __B) {
   // CHECK-LABEL: test_mm_dpwssd_avx_epi32
-  // CHECK: call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+  // CHECK: call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
   return _mm_dpwssd_avx_epi32(__S, __A, __B);
 }
 
 __m128i test_mm_dpwssds_avx_epi32(__m128i __S, __m128i __A, __m128i __B) {
   // CHECK-LABEL: test_mm_dpwssds_avx_epi32
-  // CHECK: call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+  // CHECK: call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
   return _mm_dpwssds_avx_epi32(__S, __A, __B);
 }
diff --git a/llvm/include/llvm/IR/IntrinsicsX86.td b/llvm/include/llvm/IR/IntrinsicsX86.td
index 1dd23f60c7e1e..546eed7c904a1 100644
--- a/llvm/include/llvm/IR/IntrinsicsX86.td
+++ b/llvm/include/llvm/IR/IntrinsicsX86.td
@@ -1893,29 +1893,29 @@ let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
 
   def int_x86_avx512_vpdpwssd_128 :
       ClangBuiltin<"__builtin_ia32_vpdpwssd128">,
-      DefaultAttrsIntrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty,
-                             llvm_v4i32_ty], [IntrNoMem]>;
+      DefaultAttrsIntrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v8i16_ty,
+                             llvm_v8i16_ty], [IntrNoMem]>;
   def int_x86_avx512_vpdpwssd_256 :
       ClangBuiltin<"__builtin_ia32_vpdpwssd256">,
-      DefaultAttrsIntrinsic<[llvm_v8i32_ty], [llvm_v8i32_ty, llvm_v8i32_ty,
-                             llvm_v8i32_ty], [IntrNoMem]>;
+      DefaultAttrsIntrinsic<[llvm_v8i32_ty], [llvm_v8i32_ty, llvm_v16i16_ty,
+                             llvm_v16i16_ty], [IntrNoMem]>;
   def int_x86_avx512_vpdpwssd_512 :
       ClangBuiltin<"__builtin_ia32_vpdpwssd512">,
-      DefaultAttrsIntrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, llvm_v16i32_ty,
-                             llvm_v16i32_ty], [IntrNoMem]>;
+      DefaultAttrsIntrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, llvm_v32i16_ty,
+                             llvm_v32i16_ty], [IntrNoMem]>;
 
   def int_x86_avx512_vpdpwssds_128 :
       ClangBuiltin<"__builtin_ia32_vpdpwssds128">,
-      DefaultAttrsIntrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty,
-                             llvm_v4i32_ty], [IntrNoMem]>;
+      DefaultAttrsIntrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v8i16_ty,
+                             llvm_v8i16_ty], [IntrNoMem]>;
   def int_x86_avx512_vpdpwssds_256 :
       ClangBuiltin<"__builtin_ia32_vpdpwssds256">,
-      DefaultAttrsIntrinsic<[llvm_v8i32_ty], [llvm_v8i32_ty, llvm_v8i32_ty,
-                             llvm_v8i32_ty], [IntrNoMem]>;
+      DefaultAttrsIntrinsic<[llvm_v8i32_ty], [llvm_v8i32_ty, llvm_v16i16_ty,
+                             llvm_v16i16_ty], [IntrNoMem]>;
   def int_x86_avx512_vpdpwssds_512 :
       ClangBuiltin<"__builtin_ia32_vpdpwssds512">,
-      DefaultAttrsIntrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, llvm_v16i32_ty,
-                             llvm_v16i32_ty], [IntrNoMem]>;
+      DefaultAttrsIntrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, llvm_v32i16_ty,
+                             llvm_v32i16_ty], [IntrNoMem]>;
   def int_x86_avx2_vpdpbssd_128
       : ClangBuiltin<"__builtin_ia32_vpdpbssd128">,
         DefaultAttrsIntrinsic<[llvm_v4i32_ty],
diff --git a/llvm/lib/IR/AutoUpgrade.cpp b/llvm/lib/IR/AutoUpgrade.cpp
index 58b7ddd0381e5..53ae18c4591f7 100644
--- a/llvm/lib/IR/AutoUpgrade.cpp
+++ b/llvm/lib/IR/AutoUpgrade.cpp
@@ -125,6 +125,24 @@ static bool upgradeX86MultiplyAddBytes(Function *F, Intrinsic::ID IID,
   return true;
 }
 
+// Upgrade the declaration of multipy and add words intrinsics whose input
+// arguments' types have changed to vectors of i32 to vectors of i16
+static bool upgradeX86MultiplyAddWords(Function *F, Intrinsic::ID IID,
+                                       Function *&NewFn) {
+  // check if input argument type is a vector of i16
+  Type *Arg1Type = F->getFunctionType()->getParamType(1);
+  Type *Arg2Type = F->getFunctionType()->getParamType(2);
+  if (Arg1Type->isVectorTy() &&
+      cast<VectorType>(Arg1Type)->getElementType()->isIntegerTy(16) &&
+      Arg2Type->isVectorTy() &&
+      cast<VectorType>(Arg2Type)->getElementType()->isIntegerTy(16))
+    return false;
+
+  rename(F);
+  NewFn = Intrinsic::getOrInsertDeclaration(F->getParent(), IID);
+  return true;
+}
+
 static bool upgradeX86BF16Intrinsic(Function *F, Intrinsic::ID IID,
                                     Function *&NewFn) {
   if (F->getReturnType()->getScalarType()->isBFloatTy())
@@ -590,6 +608,19 @@ static bool upgradeX86IntrinsicFunction(Function *F, StringRef Name,
                .Default(Intrinsic::not_intrinsic);
       if (ID != Intrinsic::not_intrinsic)
         return upgradeX86MultiplyAddBytes(F, ID, NewFn);
+    } else if (Name.starts_with("vpdpwssd.") ||
+               Name.starts_with("vpdpwssds.")) {
+      // Added in 21.1
+      ID = StringSwitch<Intrinsic::ID>(Name)
+               .Case("vpdpwssd.128", Intrinsic::x86_avx512_vpdpwssd_128)
+               .Case("vpdpwssd.256", Intrinsic::x86_avx512_vpdpwssd_256)
+               .Case("vpdpwssd.512", Intrinsic::x86_avx512_vpdpwssd_512)
+               .Case("vpdpwssds.128", Intrinsic::x86_avx512_vpdpwssds_128)
+               .Case("vpdpwssds.256", Intrinsic::x86_avx512_vpdpwssds_256)
+               .Case("vpdpwssds.512", Intrinsic::x86_avx512_vpdpwssds_512)
+               .Default(Intrinsic::not_intrinsic);
+      if (ID != Intrinsic::not_intrinsic)
+        return upgradeX86MultiplyAddWords(F, ID, NewFn);
     }
     return false; // No other 'x86.avx512.*'.
   }
@@ -4307,6 +4338,32 @@ static Value *upgradeX86IntrinsicCall(StringRef Name, CallBase *CI, Function *F,
 
     Value *Args[] = {CI->getArgOperand(0), CI->getArgOperand(1),
                      CI->getArgOperand(2)};
+
+    // Input arguments types were incorrectly set to vectors of i32 before but
+    // they should be vectors of i16. Insert bit cast when encountering the old
+    // types
+    if (Args[1]->getType()->isVectorTy() &&
+        cast<VectorType>(Args[1]->getType())
+            ->getElementType()
+            ->isIntegerTy(32) &&
+        Args[2]->getType()->isVectorTy() &&
+        cast<VectorType>(Args[2]->getType())
+            ->getElementType()
+            ->isIntegerTy(32)) {
+      Type *NewArgType = nullptr;
+      if (VecWidth == 128)
+        NewArgType = VectorType::get(Builder.getInt16Ty(), 8, false);
+      else if (VecWidth == 256)
+        NewArgType = VectorType::get(Builder.getInt16Ty(), 16, false);
+      else if (VecWidth == 512)
+        NewArgType = VectorType::get(Builder.getInt16Ty(), 32, false);
+      else
+        llvm_unreachable("Unexpected vector bit width");
+
+      Args[1] = Builder.CreateBitCast(Args[1], NewArgType);
+      Args[2] = Builder.CreateBitCast(Args[2], NewArgType);
+    }
+
     Rep = Builder.CreateIntrinsic(IID, Args);
     Value *PassThru = ZeroMask ? ConstantAggregateZero::get(CI->getType())
                                : CI->getArgOperand(0);
@@ -5382,6 +5439,21 @@ void llvm::UpgradeIntrinsicCall(CallBase *CI, Function *NewFn) {
     NewCall = Builder.CreateCall(NewFn, Args);
     break;
   }
+  case Intrinsic::x86_avx512_vpdpwssd_128:
+  case Intrinsic::x86_avx512_vpdpwssd_256:
+  case Intrinsic::x86_avx512_vpdpwssd_512:
+  case Intrinsic::x86_avx512_vpdpwssds_128:
+  case Intrinsic::x86_avx512_vpdpwssds_256:
+  case Intrinsic::x86_avx512_vpdpwssds_512:
+    unsigned NumElts = CI->getType()->getPrimitiveSizeInBits() / 16;
+    Value *Args[] = {CI->getArgOperand(0), CI->getArgOperand(1),
+                     CI->getArgOperand(2)};
+    Type *NewArgType = VectorType::get(Builder.getInt16Ty(), NumElts, false);
+    Args[1] = Builder.CreateBitCast(Args[1], NewArgType);
+    Args[2] = Builder.CreateBitCast(Args[2], NewArgType);
+
+    NewCall = Builder.CreateCall(NewFn, Args);
+    break;
   }
   assert(NewCall && "Should have either set this variable or returned through "
                     "the default case");
diff --git a/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp b/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp
index ceeece41782f4..f0fe71e9eb69e 100644
--- a/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp
+++ b/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp
@@ -5875,46 +5875,46 @@ struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> {
     //
     // Multiply and Add Signed Word Integers
     //   < 4 x i32> @llvm.x86.avx512.vpdpwssd.128
-    //                  (< 4 x i32>, < 4 x i32>, < 4 x i32>)
+    //                  (< 4 x i32>, < 8 x i16>, < 8 x i16>)
     //   < 8 x i32> @llvm.x86.avx512.vpdpwssd.256
-    //                  (< 8 x i32>, < 8 x i32>, < 8 x i32>)
+    //                  (< 8 x i32>, <16 x i16>, <16 x i16>)
     //   <16 x i32> @llvm.x86.avx512.vpdpwssd.512
-    //                  (<16 x i32>, <16 x i32>, <16 x i32>)
+    //                  (<16 x i32>, <32 x i16>, <32 x i16>)
     //
     // Multiply and Add Signed Word Integers With Saturation
     //   < 4 x i32> @llvm.x86.avx512.vpdpwssds.128
-    //                  (< 4 x i32>, < 4 x i32>, < 4 x i32>)
+    //                  (< 4 x i32>, < 8 x i16>, < 8 x i16>)
     //   < 8 x i32> @llvm.x86.avx512.vpdpwssds.256
-    //                  (< 8 x i32>, < 8 x i32>, < 8 x i32>)
+    //                  (< 8 x i32>, <16 x i16>, <16 x i16>)
     //   <16 x i32> @llvm.x86.avx512.vpdpwssds.512
-    //                  (<16 x i32>, <16 x i32>, <16 x i32>)
+    //                  (<16 x i32>, <32 x i16>, <32 x i16>)
     //
     // These intrinsics are auto-upgraded into non-masked forms:
     //   <4 x i32> @llvm.x86.avx512.mask.vpdpwssd.128
-    //                 (<4 x i32>, <4 x i32>, <4 x i32>, i8)
+    //                 (<4 x i32>, <8 x i16>, <8 x i16>, i8)
     //   <4 x i32> @llvm.x86.avx512.maskz.vpdpwssd.128
-    //                 (<4 x i32>, <4 x i32>, <4 x i32>, i8)
+    //                 (<4 x i32>, <8 x i16>, <8 x i16>, i8)
     //   <8 x i32> @llvm.x86.avx512.mask.vpdpwssd.256
-    //                 (<8 x i32>, <8 x i32>, <8 x i32>, i8)
+    //                 (<8 x i32>, <16 x i16>, <16 x i16>, i8)
     //   <8 x i32> @llvm.x86.avx512.maskz.vpdpwssd.256
-    //                 (<8 x i32>, <8 x i32>, <8 x i32>, i8)
+    //                 (<8 x i32>, <16 x i16>, <16 x i16>, i8)
     //   <16 x i32> @llvm.x86.avx512.mask.vpdpwssd.512
-    //                 (<16 x i32>, <16 x i32>, <16 x i32>, i16)
+    //                 (<16 x i32>, <32 x i16>, <32 x i16>, i16)
     //   <16 x i32> @llvm.x86.avx512.maskz.vpdpwssd.512
-    //                 (<16 x i32>, <16 x i32>, <16 x i32>, i16)
+    //                 (<16 x i32>, <32 x i16>, <32 x i16>, i16)
     //
     //   <4 x i32> @llvm.x86.avx512.mask.vpdpwssds.128
-    //                 (<4 x i32>, <4 x i32>, <4 x i32>, i8)
+    //                 (<4 x i32>, <8 x i16>, <8 x i16>, i8)
     //   <4 x i32> @llvm.x86.avx512.maskz.vpdpwssds.128
-    //                 (<4 x i32>, <4 x i32>, <4 x i32>, i8)
+    //                 (<4 x i32>, <8 x i16>, <8 x i16>, i8)
     //   <8 x i32> @llvm.x86.avx512.mask.vpdpwssds.256
-    //                 (<8 x i32>, <8 x i32>, <8 x i32>, i8)
+    //                 (<8 x i32>, <16 x i16>, <16 x i16>, i8)
     //   <8 x i32> @llvm.x86.avx512.maskz.vpdpwssds.256
-    //                 (<8 x i32>, <8 x i32>, <8 x i32>, i8)
+    //                 (<8 x i32>, <16 x i16>, <16 x i16>, i8)
     //   <16 x i32> @llvm.x86.avx512.mask.vpdpwssds.512
-    //                 (<16 x i32>, <16 x i32>, <16 x i32>, i16)
+    //                 (<16 x i32>, <32 x i16>, <32 x i16>, i16)
     //   <16 x i32> @llvm.x86.avx512.maskz.vpdpwssds.512
-    //                 (<16 x i32>, <16 x i32>, <16 x i32>, i16)
+    //                 (<16 x i32>, <32 x i16>, <32 x i16>, i16)
     case Intrinsic::x86_avx512_vpdpwssd_128:
     case Intrinsic::x86_avx512_vpdpwssd_256:
     case Intrinsic::x86_avx512_vpdpwssd_512:
diff --git a/llvm/test/CodeGen/X86/avx512vl_vnni-intrinsics.ll b/llvm/test/CodeGen/X86/avx512vl_vnni-intrinsics.ll
index b8ebe2a4890a1..ddf0050dbd74a 100644
--- a/llvm/test/CodeGen/X86/avx512vl_vnni-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512vl_vnni-intrinsics.ll
@@ -178,18 +178,18 @@ define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpbusds_128(<4 x i32
   ret { <4 x i32>, <4 x i32> } %res2
 }
 
-declare <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32>, <8 x i32>, <8 x i32>)
+declare <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32>, <16 x i16>, <16 x i16>)
 
-define <8 x i32>@test_int_x86_avx512_vpdpwssd_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) {
+define <8 x i32>@test_int_x86_avx512_vpdpwssd_256(<8 x i32> %x0, <16 x i16> %x1, <16 x i16> %x2) {
 ; CHECK-LABEL: test_int_x86_avx512_vpdpwssd_256:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpdpwssd %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x75,0x28,0x52,0xc2]
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %1 = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2)
+  %1 = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %x0, <16 x i16> %x1, <16 x i16> %x2)
   ret <8 x i32> %1
 }
 
-define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssd_256(<8 x i32> %x0, <8 x i32> %x1, ptr %x2p, <8 x i32> %x4, i8 %x3) {
+define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssd_256(<8 x i32> %x0, <16 x i16> %x1, ptr %x2p, <16 x i16> %x4, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask_vpdpwssd_256:
 ; X86:       # %bb.0:
 ; X86-NEXT:    vmovdqa %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xd8]
@@ -209,11 +209,11 @@ define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssd_256(<8 x i32>
 ; X64-NEXT:    vpdpwssd %ymm2, %ymm1, %ymm3 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xa9,0x52,0xda]
 ; X64-NEXT:    vmovdqa %ymm3, %ymm1 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xcb]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %x2 = load <8 x i32>, ptr %x2p
-  %1 = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2)
+  %x2 = load <16 x i16>, ptr %x2p
+  %1 = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %x0, <16 x i16> %x1, <16 x i16> %x2)
   %2 = bitcast i8 %x3 to <8 x i1>
   %3 = select <8 x i1> %2, <8 x i32> %1, <8 x i32> %x0
-  %4 = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4)
+  %4 = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %x0, <16 x i16> %x1, <16 x i16> %x4)
   %5 = bitcast i8 %x3 to <8 x i1>
   %6 = select <8 x i1> %5, <8 x i32> %4, <8 x i32> zeroinitializer
   %res1 = insertvalue { <8 x i32>, <8 x i32> } poison, <8 x i32> %3, 0
@@ -221,18 +221,18 @@ define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssd_256(<8 x i32>
   ret { <8 x i32>, <8 x i32> } %res2
 }
 
-declare <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32>, <4 x i32>, <4 x i32>)
+declare <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32>, <8 x i16>, <8 x i16>)
 
-define <4 x i32>@test_int_x86_avx512_vpdpwssd_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) {
+define <4 x i32>@test_int_x86_avx512_vpdpwssd_128(<4 x i32> %x0, <8 x i16> %x1, <8 x i16> %x2) {
 ; CHECK-LABEL: test_int_x86_avx512_vpdpwssd_128:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpdpwssd %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x75,0x08,0x52,0xc2]
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %1 = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2)
+  %1 = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %x0, <8 x i16> %x1, <8 x i16> %x2)
   ret <4 x i32> %1
 }
 
-define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssd_128(<4 x i32> %x0, <4 x i32> %x1, ptr %x2p, <4 x i32> %x4, i8 %x3) {
+define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssd_128(<4 x i32> %x0, <8 x i16> %x1, ptr %x2p, <8 x i16> %x4, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask_vpdpwssd_128:
 ; X86:       # %bb.0:
 ; X86-NEXT:    vmovdqa %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd8]
@@ -252,12 +252,12 @@ define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssd_128(<4 x i32>
 ; X64-NEXT:    vpdpwssd %xmm2, %xmm1, %xmm3 {%k1} {z} # encoding: [0x62,0xf2,0x75,0x89,0x52,0xda]
 ; X64-NEXT:    vmovdqa %xmm3, %xmm1 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xcb]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %x2 = load <4 x i32>, ptr %x2p
-  %1 = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2)
+  %x2 = load <8 x i16>, ptr %x2p
+  %1 = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %x0, <8 x i16> %x1, <8 x i16> %x2)
   %2 = bitcast i8 %x3 to <8 x i1>
   %extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %3 = select <4 x i1> %extract, <4 x i32> %1, <4 x i32> %x0
-  %4 = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4)
+  %4 = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %x0, <8 x i16> %x1, <8 x i16> %x4)
   %5 = bitcast i8 %x3 to <8 x i1>
   %extract1 = shufflevector <8 x i1> %5, <8 x i1> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %6 = select <4 x i1> %extract1, <4 x i32> %4, <4 x i32> zeroinitializer
@@ -266,18 +266,18 @@ define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssd_128(<4 x i32>
   ret { <4 x i32>, <4 x i32> } %res2
 }
 
-declare <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32>, <8 x i32>, <8 x i32>)
+declare <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32>, <16 x i16>, <16 x i16>)
 
-define <8 x i32>@test_int_x86_avx512_vpdpwssds_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) {
+define <8 x i32>@test_int_x86_avx512_vpdpwssds_256(<8 x i32> %x0, <16 x i16> %x1, <16 x i16> %x2) {
 ; CHECK-LABEL: test_int_x86_avx512_vpdpwssds_256:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpdpwssds %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x75,0x28,0x53,0xc2]
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %1 = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2)
+  %1 = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %x0, <16 x i16> %x1, <16 x i16> %x2)
   ret <8 x i32> %1
 }
 
-define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssds_256(<8 x i32> %x0, <8 x i32> %x1, ptr %x2p, <8 x i32> %x4, i8 %x3) {
+define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssds_256(<8 x i32> %x0, <16 x i16> %x1, ptr %x2p, <16 x i16> %x4, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask_vpdpwssds_256:
 ; X86:       # %bb.0:
 ; X86-NEXT:    vmovdqa %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xd8]
@@ -297,11 +297,11 @@ define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssds_256(<8 x i32
 ; X64-NEXT:    vpdpwssds %ymm2, %ymm1, %ymm3 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xa9,0x53,0xda]
 ; X64-NEXT:    vmovdqa %ymm3, %ymm1 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xcb]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %x2 = load <8 x i32>, ptr %x2p
-  %1 = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2)
+  %x2 = load <16 x i16>, ptr %x2p
+  %1 = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %x0, <16 x i16> %x1, <16 x i16> %x2)
   %2 = bitcast i8 %x3 to <8 x i1>
   %3 = select <8 x i1> %2, <8 x i32> %1, <8 x i32> %x0
-  %4 = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4)
+  %4 = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %x0, <16 x i16> %x1, <16 x i16> %x4)
   %5 = bitcast i8 %x3 to <8 x i1>
   %6 = select <8 x i1> %5, <8 x i32> %4, <8 x i32> zeroinitializer
   %res1 = insertvalue { <8 x i32>, <8 x i32> } poison, <8 x i32> %3, 0
@@ -309,9 +309,9 @@ define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssds_256(<8 x i32
   ret { <8 x i32>, <8 x i32> } %res2
 }
 
-declare <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32>, <4 x i32>, <4 x i32>)
+declare <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32>, <8 x i16>, <8 x i16>)
 
-define <4 x i32>@test_int_x86_avx512_vpdpwssds_128(<4 x i32> %x0, <4 x i32> %x1, ptr %x2p) {
+define <4 x i32>@test_int_x86_avx512_vpdpwssds_128(<4 x i32> %x0, <8 x i16> %x1, ptr %x2p) {
 ; X86-LABEL: test_int_x86_avx512_vpdpwssds_128:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
@@ -322,12 +322,12 @@ define <4 x i32>@test_int_x86_avx512_vpdpwssds_128(<4 x i32> %x0, <4 x i32> %x1,
 ; X64:       # %bb.0:
 ; X64-NEXT:    vpdpwssds (%rdi), %xmm1, %xmm0 # encoding: [0x62,0xf2,0x75,0x08,0x53,0x07]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %x2 = load <4 x i32>, ptr %x2p
-  %1 = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2)
+  %x2 = load <8 x i16>, ptr %x2p
+  %1 = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %x0, <8 x i16> %x1, <8 x i16> %x2)
   ret <4 x i32> %1
 }
 
-define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssds_128(<4 x i32> %x0, <4 x i32> %x1, ptr %x2p, <4 x i32> %x4, i8 %x3) {
+define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssds_128(<4 x i32> %x0, <8 x i16> %x1, ptr %x2p, <8 x i16> %x4, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask_vpdpwssds_128:
 ; X86:       # %bb.0:
 ; X86-NEXT:    vmovdqa %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd8]
@@ -347,12 +347,12 @@ define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssds_128(<4 x i32
 ; X64-NEXT:    vpdpwssds %xmm2, %xmm1, %xmm3 {%k1} {z} # encoding: [0x62,0xf2,0x75,0x89,0x53,0xda]
 ; X64-NEXT:    vmovdqa %xmm3, %xmm1 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xcb]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %x2 = load <4 x i32>, ptr %x2p
-  %1 = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2)
+  %x2 = load <8 x i16>, ptr %x2p
+  %1 = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %x0, <8 x i16> %x1, <8 x i16> %x2)
   %2 = bitcast i8 %x3 to <8 x i1>
   %extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %3 = select <4 x i1> %extract, <4 x i32> %1, <4 x i32> %x0
-  %4 = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4)
+  %4 = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %x0, <8 x i16> %x1, <8 x i16> %x4)
   %5 = bitcast i8 %x3 to <8 x i1>
   %extract1 = shufflevector <8 x i1> %5, <8 x i1> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %6 = select <4 x i1> %extract1, <4 x i32> %4, <4 x i32> zeroinitializer
diff --git a/llvm/test/CodeGen/X86/avx512vnni-intrinsics-upgrade.ll b/llvm/test/CodeGen/X86/avx512vnni-intrinsics-upgrade.ll
index 63ff88a7fa4ae..35363e99ef4c8 100644
--- a/llvm/test/CodeGen/X86/avx512vnni-intrinsics-upgrade.ll
+++ b/llvm/test/CodeGen/X86/avx512vnni-intrinsics-upgrade.ll
@@ -102,11 +102,22 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_maskz_vpdpbusds_512(<16 x
   ret { <16 x i32>, <16 x i32> } %res3
 }
 
+declare <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
+
+define <16 x i32>@test_int_x86_avx512_vpdpwssd(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
+; CHECK-LABEL: test_int_x86_avx512_vpdpwssd:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpdpwssd %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0x52,0xc2]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %res = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
+  ret <16 x i32> %res
+}
+
 declare <16 x i32> @llvm.x86.avx512.mask.vpdpwssd.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
 declare <16 x i32> @llvm.x86.avx512.maskz.vpdpwssd.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
 
-define <16 x i32>@test_int_x86_avx512_vpdpwssd_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
-; CHECK-LABEL: test_int_x86_avx512_vpdpwssd_512:
+define <16 x i32>@test_int_x86_avx512_mask_vpdpwssd_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
+; CHECK-LABEL: test_int_x86_avx512_mask_vpdpwssd_512:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpdpwssd %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0x52,0xc2]
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
@@ -114,8 +125,8 @@ define <16 x i32>@test_int_x86_avx512_vpdpwssd_512(<16 x i32> %x0, <16 x i32> %x
   ret <16 x i32> %res
 }
 
-define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssd_512(<16 x i32> %x0, <16 x i32> %x1, ptr %x2p, <16 x i32> %x4, i16 %x3) {
-; X86-LABEL: test_int_x86_avx512_mask_vpdpwssd_512:
+define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_maskz_vpdpwssd_512(<16 x i32> %x0, <16 x i32> %x1, ptr %x2p, <16 x i32> %x4, i16 %x3) {
+; X86-LABEL: test_int_x86_avx512_maskz_vpdpwssd_512:
 ; X86:       # %bb.0:
 ; X86-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
@@ -125,7 +136,7 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssd_512(<16 x i
 ; X86-NEXT:    vmovdqa64 %zmm3, %zmm1 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xcb]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
-; X64-LABEL: test_int_x86_avx512_mask_vpdpwssd_512:
+; X64-LABEL: test_int_x86_avx512_maskz_vpdpwssd_512:
 ; X64:       # %bb.0:
 ; X64-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
@@ -141,20 +152,31 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssd_512(<16 x i
   ret { <16 x i32>, <16 x i32> } %res3
 }
 
-declare <16 x i32> @llvm.x86.avx512.mask.vpdpwssds.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
-declare <16 x i32> @llvm.x86.avx512.maskz.vpdpwssds.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
+declare <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
 
 define <16 x i32>@test_int_x86_avx512_vpdpwssds_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
 ; CHECK-LABEL: test_int_x86_avx512_vpdpwssds_512:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpdpwssds %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0x53,0xc2]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %res = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
+  ret <16 x i32> %res
+}
+
+declare <16 x i32> @llvm.x86.avx512.mask.vpdpwssds.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
+declare <16 x i32> @llvm.x86.avx512.maskz.vpdpwssds.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
+
+define <16 x i32>@test_int_x86_avx512_mask_vpdpwssds_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
+; CHECK-LABEL: test_int_x86_avx512_mask_vpdpwssds_512:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpdpwssds %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0x53,0xc2]
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
   %res = call <16 x i32> @llvm.x86.avx512.mask.vpdpwssds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 -1)
   ret <16 x i32> %res
 }
 
-define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssds_512(<16 x i32> %x0, <16 x i32> %x1, ptr %x2p, <16 x i32> %x4, i16 %x3) {
-; X86-LABEL: test_int_x86_avx512_mask_vpdpwssds_512:
+define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_maskz_vpdpwssds_512(<16 x i32> %x0, <16 x i32> %x1, ptr %x2p, <16 x i32> %x4, i16 %x3) {
+; X86-LABEL: test_int_x86_avx512_maskz_vpdpwssds_512:
 ; X86:       # %bb.0:
 ; X86-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
@@ -164,7 +186,7 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssds_512(<16 x
 ; X86-NEXT:    vmovdqa64 %zmm3, %zmm1 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xcb]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
-; X64-LABEL: test_int_x86_avx512_mask_vpdpwssds_512:
+; X64-LABEL: test_int_x86_avx512_maskz_vpdpwssds_512:
 ; X64:       # %bb.0:
 ; X64-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
diff --git a/llvm/test/CodeGen/X86/avx512vnni-intrinsics.ll b/llvm/test/CodeGen/X86/avx512vnni-intrinsics.ll
index 60d0298e057f3..e97b8a5c5503f 100644
--- a/llvm/test/CodeGen/X86/avx512vnni-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512vnni-intrinsics.ll
@@ -86,18 +86,18 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpbusds_512(<16 x
   ret { <16 x i32>, <16 x i32> } %res2
 }
 
-declare <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32>, <16 x i32>, <16 x i32>)
+declare <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32>, <32 x i16>, <32 x i16>)
 
-define <16 x i32>@test_int_x86_avx512_vpdpwssd_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
+define <16 x i32>@test_int_x86_avx512_vpdpwssd_512(<16 x i32> %x0, <32 x i16> %x1, <32 x i16> %x2) {
 ; CHECK-LABEL: test_int_x86_avx512_vpdpwssd_512:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpdpwssd %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0x52,0xc2]
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %1 = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
+  %1 = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %x0, <32 x i16> %x1, <32 x i16> %x2)
   ret <16 x i32> %1
 }
 
-define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssd_512(<16 x i32> %x0, <16 x i32> %x1, ptr %x2p, <16 x i32> %x4, i16 %x3) {
+define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssd_512(<16 x i32> %x0, <32 x i16> %x1, ptr %x2p, <32 x i16> %x4, i16 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask_vpdpwssd_512:
 ; X86:       # %bb.0:
 ; X86-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
@@ -116,11 +116,11 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssd_512(<16 x i
 ; X64-NEXT:    vpdpwssd %zmm2, %zmm1, %zmm3 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x52,0xda]
 ; X64-NEXT:    vmovdqa64 %zmm3, %zmm1 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xcb]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %x2 = load <16 x i32>, ptr %x2p
-  %1 = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
+  %x2 = load <32 x i16>, ptr %x2p
+  %1 = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %x0, <32 x i16> %x1, <32 x i16> %x2)
   %2 = bitcast i16 %x3 to <16 x i1>
   %3 = select <16 x i1> %2, <16 x i32> %1, <16 x i32> %x0
-  %4 = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x4)
+  %4 = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> %x0, <32 x i16> %x1, <32 x i16> %x4)
   %5 = bitcast i16 %x3 to <16 x i1>
   %6 = select <16 x i1> %5, <16 x i32> %4, <16 x i32> zeroinitializer
   %res1 = insertvalue { <16 x i32>, <16 x i32> } poison, <16 x i32> %3, 0
@@ -128,18 +128,18 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssd_512(<16 x i
   ret { <16 x i32>, <16 x i32> } %res2
 }
 
-declare <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32>, <16 x i32>, <16 x i32>)
+declare <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32>, <32 x i16>, <32 x i16>)
 
-define <16 x i32>@test_int_x86_avx512_ask_vpdpwssds_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
+define <16 x i32>@test_int_x86_avx512_ask_vpdpwssds_512(<16 x i32> %x0, <32 x i16> %x1, <32 x i16> %x2) {
 ; CHECK-LABEL: test_int_x86_avx512_ask_vpdpwssds_512:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpdpwssds %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0x53,0xc2]
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %1 = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
+  %1 = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %x0, <32 x i16> %x1, <32 x i16> %x2)
   ret <16 x i32> %1
 }
 
-define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssds_512(<16 x i32> %x0, <16 x i32> %x1, ptr %x2p, <16 x i32> %x4, i16 %x3) {
+define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssds_512(<16 x i32> %x0, <32 x i16> %x1, ptr %x2p, <32 x i16> %x4, i16 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask_vpdpwssds_512:
 ; X86:       # %bb.0:
 ; X86-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
@@ -158,11 +158,11 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssds_512(<16 x
 ; X64-NEXT:    vpdpwssds %zmm2, %zmm1, %zmm3 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0x53,0xda]
 ; X64-NEXT:    vmovdqa64 %zmm3, %zmm1 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xcb]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %x2 = load <16 x i32>, ptr %x2p
-  %1 = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
+  %x2 = load <32 x i16>, ptr %x2p
+  %1 = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %x0, <32 x i16> %x1, <32 x i16> %x2)
   %2 = bitcast i16 %x3 to <16 x i1>
   %3 = select <16 x i1> %2, <16 x i32> %1, <16 x i32> %x0
-  %4 = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x4)
+  %4 = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> %x0, <32 x i16> %x1, <32 x i16> %x4)
   %5 = bitcast i16 %x3 to <16 x i1>
   %6 = select <16 x i1> %5, <16 x i32> %4, <16 x i32> zeroinitializer
   %res1 = insertvalue { <16 x i32>, <16 x i32> } poison, <16 x i32> %3, 0
diff --git a/llvm/test/CodeGen/X86/avx_vnni-intrinsics-upgrade.ll b/llvm/test/CodeGen/X86/avx_vnni-intrinsics-upgrade.ll
index 0f4a4f27b9715..f359ecef8ceb3 100644
--- a/llvm/test/CodeGen/X86/avx_vnni-intrinsics-upgrade.ll
+++ b/llvm/test/CodeGen/X86/avx_vnni-intrinsics-upgrade.ll
@@ -45,3 +45,47 @@ define <8 x i32>@test_int_x86_avx_vpdpbusds_256(<8 x i32> %x0, <8 x i32> %x1, <8
     %res = call <8 x i32> @llvm.x86.avx512.vpdpbusds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2)
     ret <8 x i32> %res
 }
+
+declare <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32>, <4 x i32>, <4 x i32>)
+
+define <4 x i32>@test_int_x86_avx_vpdpwssd_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) {
+; CHECK-LABEL: test_int_x86_avx_vpdpwssd_128:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    {vex} vpdpwssd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0x52,0xc2]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+    %res = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2)
+    ret <4 x i32> %res
+}
+
+declare <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32>, <8 x i32>, <8 x i32>)
+
+define <8 x i32>@test_int_x86_avx_vpdpwssd_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) {
+; CHECK-LABEL: test_int_x86_avx_vpdpwssd_256:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    {vex} vpdpwssd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0x52,0xc2]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+    %res = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2)
+    ret <8 x i32> %res
+}
+
+declare <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32>, <4 x i32>, <4 x i32>)
+
+define <4 x i32>@test_int_x86_avx_vpdpwssds_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) {
+; CHECK-LABEL: test_int_x86_avx_vpdpwssds_128:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    {vex} vpdpwssds %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0x53,0xc2]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+    %res = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2)
+    ret <4 x i32> %res
+}
+
+declare <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32>, <8 x i32>, <8 x i32>)
+
+define <8 x i32>@test_int_x86_avx_vpdpwssds_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) {
+; CHECK-LABEL: test_int_x86_avx_vpdpwssds_256:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    {vex} vpdpwssds %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0x53,0xc2]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+    %res = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2)
+    ret <8 x i32> %res
+}
diff --git a/llvm/test/CodeGen/X86/avx_vnni-intrinsics.ll b/llvm/test/CodeGen/X86/avx_vnni-intrinsics.ll
index de8b2a41bf8c8..5748a426c76c3 100644
--- a/llvm/test/CodeGen/X86/avx_vnni-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx_vnni-intrinsics.ll
@@ -68,9 +68,9 @@ define <4 x i32>@test_int_x86_avx_vpdpbusds_128(<4 x i32> %x0, <16 x i8> %x1, <1
   ret <4 x i32> %res
 }
 
-declare <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32>, <8 x i32>, <8 x i32>)
+declare <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32>, <16 x i16>, <16 x i16>)
 
-define <8 x i32>@test_int_x86_avx_vpdpwssd_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) {
+define <8 x i32>@test_int_x86_avx_vpdpwssd_256(<8 x i32> %x0, <16 x i16> %x1, <16 x i16> %x2) {
 ; AVXVNNI-LABEL: test_int_x86_avx_vpdpwssd_256:
 ; AVXVNNI:       # %bb.0:
 ; AVXVNNI-NEXT:    {vex} vpdpwssd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0x52,0xc2]
@@ -80,13 +80,13 @@ define <8 x i32>@test_int_x86_avx_vpdpwssd_256(<8 x i32> %x0, <8 x i32> %x1, <8
 ; AVX512VNNI:       # %bb.0:
 ; AVX512VNNI-NEXT:    {vex} vpdpwssd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0x52,0xc2]
 ; AVX512VNNI-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2)
+  %res = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %x0, <16 x i16> %x1, <16 x i16> %x2)
   ret <8 x i32> %res
 }
 
-declare <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32>, <4 x i32>, <4 x i32>)
+declare <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32>, <8 x i16>, <8 x i16>)
 
-define <4 x i32>@test_int_x86_avx_vpdpwssd_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) {
+define <4 x i32>@test_int_x86_avx_vpdpwssd_128(<4 x i32> %x0, <8 x i16> %x1, <8 x i16> %x2) {
 ; AVXVNNI-LABEL: test_int_x86_avx_vpdpwssd_128:
 ; AVXVNNI:       # %bb.0:
 ; AVXVNNI-NEXT:    {vex} vpdpwssd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0x52,0xc2]
@@ -96,13 +96,13 @@ define <4 x i32>@test_int_x86_avx_vpdpwssd_128(<4 x i32> %x0, <4 x i32> %x1, <4
 ; AVX512VNNI:       # %bb.0:
 ; AVX512VNNI-NEXT:    {vex} vpdpwssd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0x52,0xc2]
 ; AVX512VNNI-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2)
+  %res = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %x0, <8 x i16> %x1, <8 x i16> %x2)
   ret <4 x i32> %res
 }
 
-declare <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32>, <8 x i32>, <8 x i32>)
+declare <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32>, <16 x i16>, <16 x i16>)
 
-define <8 x i32>@test_int_x86_avx_vpdpwssds_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) {
+define <8 x i32>@test_int_x86_avx_vpdpwssds_256(<8 x i32> %x0, <16 x i16> %x1, <16 x i16> %x2) {
 ; AVXVNNI-LABEL: test_int_x86_avx_vpdpwssds_256:
 ; AVXVNNI:       # %bb.0:
 ; AVXVNNI-NEXT:    {vex} vpdpwssds %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0x53,0xc2]
@@ -112,13 +112,13 @@ define <8 x i32>@test_int_x86_avx_vpdpwssds_256(<8 x i32> %x0, <8 x i32> %x1, <8
 ; AVX512VNNI:       # %bb.0:
 ; AVX512VNNI-NEXT:    {vex} vpdpwssds %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0x53,0xc2]
 ; AVX512VNNI-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2)
+  %res = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %x0, <16 x i16> %x1, <16 x i16> %x2)
   ret <8 x i32> %res
 }
 
-declare <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32>, <4 x i32>, <4 x i32>)
+declare <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32>, <8 x i16>, <8 x i16>)
 
-define <4 x i32>@test_int_x86_avx_vpdpwssds_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) {
+define <4 x i32>@test_int_x86_avx_vpdpwssds_128(<4 x i32> %x0, <8 x i16> %x1, <8 x i16> %x2) {
 ; AVXVNNI-LABEL: test_int_x86_avx_vpdpwssds_128:
 ; AVXVNNI:       # %bb.0:
 ; AVXVNNI-NEXT:    {vex} vpdpwssds %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0x53,0xc2]
@@ -128,6 +128,6 @@ define <4 x i32>@test_int_x86_avx_vpdpwssds_128(<4 x i32> %x0, <4 x i32> %x1, <4
 ; AVX512VNNI:       # %bb.0:
 ; AVX512VNNI-NEXT:    {vex} vpdpwssds %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0x53,0xc2]
 ; AVX512VNNI-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2)
+  %res = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %x0, <8 x i16> %x1, <8 x i16> %x2)
   ret <4 x i32> %res
 }
diff --git a/llvm/test/CodeGen/X86/stack-folding-int-avxvnni.ll b/llvm/test/CodeGen/X86/stack-folding-int-avxvnni.ll
index cd576b19f8766..345fa0efa42df 100644
--- a/llvm/test/CodeGen/X86/stack-folding-int-avxvnni.ll
+++ b/llvm/test/CodeGen/X86/stack-folding-int-avxvnni.ll
@@ -4,16 +4,16 @@
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-unknown-unknown"
 
-declare <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32>, <4 x i32>, <4 x i32>)
-declare <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32>, <8 x i32>, <8 x i32>)
-declare <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32>, <4 x i32>, <4 x i32>)
-declare <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32>, <8 x i32>, <8 x i32>)
+declare <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32>, <8 x i16>, <8 x i16>)
+declare <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32>, <16 x i16>, <16 x i16>)
+declare <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32>, <8 x i16>, <8 x i16>)
+declare <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32>, <16 x i16>, <16 x i16>)
 declare <4 x i32> @llvm.x86.avx512.vpdpbusd.128(<4 x i32>, <16 x i8>, <16 x i8>)
 declare <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32>, <32 x i8>, <32 x i8>)
 declare <4 x i32> @llvm.x86.avx512.vpdpbusds.128(<4 x i32>, <16 x i8>, <16 x i8>)
 declare <8 x i32> @llvm.x86.avx512.vpdpbusds.256(<8 x i32>, <32 x i8>, <32 x i8>)
 
-define <4 x i32> @stack_fold_vpdpwssd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) {
+define <4 x i32> @stack_fold_vpdpwssd(<4 x i32> %a0, <8 x i16> %a1, <8 x i16> %a2) {
 ; CHECK-LABEL: stack_fold_vpdpwssd:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
@@ -23,11 +23,11 @@ define <4 x i32> @stack_fold_vpdpwssd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a
 ; CHECK-NEXT:    {vex} vpdpwssd {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    retq
   %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
-  %2 = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2)
+  %2 = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %a0, <8 x i16> %a1, <8 x i16> %a2)
   ret <4 x i32> %2
 }
 
-define <4 x i32> @stack_fold_vpdpwssd_commuted(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) {
+define <4 x i32> @stack_fold_vpdpwssd_commuted(<4 x i32> %a0, <8 x i16> %a1, <8 x i16> %a2) {
 ; CHECK-LABEL: stack_fold_vpdpwssd_commuted:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
@@ -37,11 +37,11 @@ define <4 x i32> @stack_fold_vpdpwssd_commuted(<4 x i32> %a0, <4 x i32> %a1, <4
 ; CHECK-NEXT:    {vex} vpdpwssd {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    retq
   %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
-  %2 = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %a0, <4 x i32> %a2, <4 x i32> %a1)
+  %2 = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %a0, <8 x i16> %a2, <8 x i16> %a1)
   ret <4 x i32> %2
 }
 
-define <8 x i32> @stack_fold_vpdpwssd_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) {
+define <8 x i32> @stack_fold_vpdpwssd_256(<8 x i32> %a0, <16 x i16> %a1, <16 x i16> %a2) {
 ; CHECK-LABEL: stack_fold_vpdpwssd_256:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -51,11 +51,11 @@ define <8 x i32> @stack_fold_vpdpwssd_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32
 ; CHECK-NEXT:    {vex} vpdpwssd {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload
 ; CHECK-NEXT:    retq
   %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
-  %2 = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2)
+  %2 = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %a0, <16 x i16> %a1, <16 x i16> %a2)
   ret <8 x i32> %2
 }
 
-define <8 x i32> @stack_fold_vpdpwssd_256_commuted(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) {
+define <8 x i32> @stack_fold_vpdpwssd_256_commuted(<8 x i32> %a0, <16 x i16> %a1, <16 x i16> %a2) {
 ; CHECK-LABEL: stack_fold_vpdpwssd_256_commuted:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -65,11 +65,11 @@ define <8 x i32> @stack_fold_vpdpwssd_256_commuted(<8 x i32> %a0, <8 x i32> %a1,
 ; CHECK-NEXT:    {vex} vpdpwssd {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload
 ; CHECK-NEXT:    retq
   %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
-  %2 = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %a0, <8 x i32> %a2, <8 x i32> %a1)
+  %2 = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %a0, <16 x i16> %a2, <16 x i16> %a1)
   ret <8 x i32> %2
 }
 
-define <4 x i32> @stack_fold_vpdpwssds(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) {
+define <4 x i32> @stack_fold_vpdpwssds(<4 x i32> %a0, <8 x i16> %a1, <8 x i16> %a2) {
 ; CHECK-LABEL: stack_fold_vpdpwssds:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
@@ -79,11 +79,11 @@ define <4 x i32> @stack_fold_vpdpwssds(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %
 ; CHECK-NEXT:    {vex} vpdpwssds {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    retq
   %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
-  %2 = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2)
+  %2 = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %a0, <8 x i16> %a1, <8 x i16> %a2)
   ret <4 x i32> %2
 }
 
-define <4 x i32> @stack_fold_vpdpwssds_commuted(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) {
+define <4 x i32> @stack_fold_vpdpwssds_commuted(<4 x i32> %a0, <8 x i16> %a1, <8 x i16> %a2) {
 ; CHECK-LABEL: stack_fold_vpdpwssds_commuted:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
@@ -93,11 +93,11 @@ define <4 x i32> @stack_fold_vpdpwssds_commuted(<4 x i32> %a0, <4 x i32> %a1, <4
 ; CHECK-NEXT:    {vex} vpdpwssds {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    retq
   %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
-  %2 = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %a0, <4 x i32> %a2, <4 x i32> %a1)
+  %2 = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %a0, <8 x i16> %a2, <8 x i16> %a1)
   ret <4 x i32> %2
 }
 
-define <8 x i32> @stack_fold_vpdpwssds_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) {
+define <8 x i32> @stack_fold_vpdpwssds_256(<8 x i32> %a0, <16 x i16> %a1, <16 x i16> %a2) {
 ; CHECK-LABEL: stack_fold_vpdpwssds_256:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -107,11 +107,11 @@ define <8 x i32> @stack_fold_vpdpwssds_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i3
 ; CHECK-NEXT:    {vex} vpdpwssds {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload
 ; CHECK-NEXT:    retq
   %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
-  %2 = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2)
+  %2 = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %a0, <16 x i16> %a1, <16 x i16> %a2)
   ret <8 x i32> %2
 }
 
-define <8 x i32> @stack_fold_vpdpwssds_256_commuted(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) {
+define <8 x i32> @stack_fold_vpdpwssds_256_commuted(<8 x i32> %a0, <16 x i16> %a1, <16 x i16> %a2) {
 ; CHECK-LABEL: stack_fold_vpdpwssds_256_commuted:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -121,7 +121,7 @@ define <8 x i32> @stack_fold_vpdpwssds_256_commuted(<8 x i32> %a0, <8 x i32> %a1
 ; CHECK-NEXT:    {vex} vpdpwssds {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload
 ; CHECK-NEXT:    retq
   %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
-  %2 = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %a0, <8 x i32> %a2, <8 x i32> %a1)
+  %2 = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %a0, <16 x i16> %a2, <16 x i16> %a1)
   ret <8 x i32> %2
 }
 
diff --git a/llvm/test/Instrumentation/MemorySanitizer/X86/avx512vl_vnni-intrinsics-upgrade.ll b/llvm/test/Instrumentation/MemorySanitizer/X86/avx512vl_vnni-intrinsics-upgrade.ll
index 5e937485ff282..d0daea5e68fea 100644
--- a/llvm/test/Instrumentation/MemorySanitizer/X86/avx512vl_vnni-intrinsics-upgrade.ll
+++ b/llvm/test/Instrumentation/MemorySanitizer/X86/avx512vl_vnni-intrinsics-upgrade.ll
@@ -528,10 +528,10 @@ define <8 x i32>@test_int_x86_avx512_vpdpwssd_256(<8 x i32> %x0, <8 x i32> %x1,
 ; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
 ; CHECK-NEXT:    [[TMP21:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
-; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <8 x i32> [[TMP2]] to <16 x i16>
+; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <8 x i32> [[TMP3]] to <16 x i16>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne <16 x i16> [[TMP6]], zeroinitializer
 ; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <16 x i16> [[TMP7]], zeroinitializer
 ; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne <16 x i16> [[TMP22]], zeroinitializer
@@ -546,7 +546,7 @@ define <8 x i32>@test_int_x86_avx512_vpdpwssd_256(<8 x i32> %x0, <8 x i32> %x1,
 ; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <8 x i32> [[TMP18]], zeroinitializer
 ; CHECK-NEXT:    [[TMP20:%.*]] = sext <8 x i1> [[TMP19]] to <8 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[TMP20]], [[TMP21]]
-; CHECK-NEXT:    [[TMP4:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP4:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> [[X0]], <16 x i16> [[TMP22]], <16 x i16> [[TMP5]])
 ; CHECK-NEXT:    store <8 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <8 x i32> [[TMP4]]
 ;
@@ -574,10 +574,10 @@ define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssd_256(<8 x i32>
 ; CHECK-NEXT:    [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080
 ; CHECK-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load <8 x i32>, ptr [[TMP10]], align 32
-; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
-; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP31:%.*]] = bitcast <8 x i32> [[TMP3]] to <16 x i16>
+; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP32:%.*]] = bitcast <8 x i32> [[_MSLD]] to <16 x i16>
+; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP33:%.*]] = icmp ne <16 x i16> [[TMP31]], zeroinitializer
 ; CHECK-NEXT:    [[TMP34:%.*]] = icmp ne <16 x i16> [[TMP32]], zeroinitializer
 ; CHECK-NEXT:    [[TMP35:%.*]] = icmp ne <16 x i16> [[TMP29]], zeroinitializer
@@ -592,7 +592,7 @@ define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssd_256(<8 x i32>
 ; CHECK-NEXT:    [[TMP61:%.*]] = icmp ne <8 x i32> [[TMP60]], zeroinitializer
 ; CHECK-NEXT:    [[TMP62:%.*]] = sext <8 x i1> [[TMP61]] to <8 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[TMP62]], [[TMP2]]
-; CHECK-NEXT:    [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> [[X0]], <16 x i16> [[TMP29]], <16 x i16> [[TMP30]])
 ; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i8 [[X3]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP14:%.*]] = select <8 x i1> [[TMP13]], <8 x i32> [[_MSPROP1]], <8 x i32> [[TMP2]]
@@ -601,10 +601,10 @@ define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssd_256(<8 x i32>
 ; CHECK-NEXT:    [[TMP17:%.*]] = or <8 x i32> [[TMP16]], [[TMP2]]
 ; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <8 x i1> [[TMP12]], <8 x i32> [[TMP17]], <8 x i32> [[TMP14]]
 ; CHECK-NEXT:    [[TMP18:%.*]] = select <8 x i1> [[TMP13]], <8 x i32> [[TMP11]], <8 x i32> [[X0]]
-; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
-; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <8 x i32> [[X4]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP39:%.*]] = bitcast <8 x i32> [[TMP3]] to <16 x i16>
+; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP40:%.*]] = bitcast <8 x i32> [[TMP5]] to <16 x i16>
+; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <8 x i32> [[X4]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP41:%.*]] = icmp ne <16 x i16> [[TMP39]], zeroinitializer
 ; CHECK-NEXT:    [[TMP42:%.*]] = icmp ne <16 x i16> [[TMP40]], zeroinitializer
 ; CHECK-NEXT:    [[TMP43:%.*]] = icmp ne <16 x i16> [[TMP37]], zeroinitializer
@@ -619,7 +619,7 @@ define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssd_256(<8 x i32>
 ; CHECK-NEXT:    [[TMP52:%.*]] = icmp ne <8 x i32> [[TMP51]], zeroinitializer
 ; CHECK-NEXT:    [[TMP53:%.*]] = sext <8 x i1> [[TMP52]] to <8 x i32>
 ; CHECK-NEXT:    [[_MSPROP3:%.*]] = or <8 x i32> [[TMP53]], [[TMP2]]
-; CHECK-NEXT:    [[TMP19:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X4]])
+; CHECK-NEXT:    [[TMP19:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> [[X0]], <16 x i16> [[TMP37]], <16 x i16> [[TMP38]])
 ; CHECK-NEXT:    [[TMP20:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP21:%.*]] = bitcast i8 [[X3]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP22:%.*]] = select <8 x i1> [[TMP21]], <8 x i32> [[_MSPROP3]], <8 x i32> zeroinitializer
@@ -653,10 +653,10 @@ define <4 x i32>@test_int_x86_avx512_vpdpwssd_128(<4 x i32> %x0, <4 x i32> %x1,
 ; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
 ; CHECK-NEXT:    [[TMP21:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
-; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <4 x i32> [[TMP2]] to <8 x i16>
+; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <4 x i32> [[TMP3]] to <8 x i16>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne <8 x i16> [[TMP6]], zeroinitializer
 ; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <8 x i16> [[TMP7]], zeroinitializer
 ; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne <8 x i16> [[TMP22]], zeroinitializer
@@ -671,7 +671,7 @@ define <4 x i32>@test_int_x86_avx512_vpdpwssd_128(<4 x i32> %x0, <4 x i32> %x1,
 ; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <4 x i32> [[TMP18]], zeroinitializer
 ; CHECK-NEXT:    [[TMP20:%.*]] = sext <4 x i1> [[TMP19]] to <4 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[TMP20]], [[TMP21]]
-; CHECK-NEXT:    [[TMP4:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP4:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> [[X0]], <8 x i16> [[TMP22]], <8 x i16> [[TMP5]])
 ; CHECK-NEXT:    store <4 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <4 x i32> [[TMP4]]
 ;
@@ -699,10 +699,10 @@ define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssd_128(<4 x i32>
 ; CHECK-NEXT:    [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080
 ; CHECK-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load <4 x i32>, ptr [[TMP10]], align 16
-; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
-; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP31:%.*]] = bitcast <4 x i32> [[TMP3]] to <8 x i16>
+; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP32:%.*]] = bitcast <4 x i32> [[_MSLD]] to <8 x i16>
+; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP33:%.*]] = icmp ne <8 x i16> [[TMP31]], zeroinitializer
 ; CHECK-NEXT:    [[TMP34:%.*]] = icmp ne <8 x i16> [[TMP32]], zeroinitializer
 ; CHECK-NEXT:    [[TMP35:%.*]] = icmp ne <8 x i16> [[TMP29]], zeroinitializer
@@ -717,7 +717,7 @@ define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssd_128(<4 x i32>
 ; CHECK-NEXT:    [[TMP61:%.*]] = icmp ne <4 x i32> [[TMP60]], zeroinitializer
 ; CHECK-NEXT:    [[TMP62:%.*]] = sext <4 x i1> [[TMP61]] to <4 x i32>
 ; CHECK-NEXT:    [[_MSPROP2:%.*]] = or <4 x i32> [[TMP62]], [[TMP2]]
-; CHECK-NEXT:    [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> [[X0]], <8 x i16> [[TMP29]], <8 x i16> [[TMP30]])
 ; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i8 [[X3]] to <8 x i1>
 ; CHECK-NEXT:    [[_MSPROP3:%.*]] = shufflevector <8 x i1> [[TMP12]], <8 x i1> [[TMP12]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
@@ -728,10 +728,10 @@ define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssd_128(<4 x i32>
 ; CHECK-NEXT:    [[TMP17:%.*]] = or <4 x i32> [[TMP16]], [[TMP2]]
 ; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <4 x i1> [[_MSPROP3]], <4 x i32> [[TMP17]], <4 x i32> [[TMP14]]
 ; CHECK-NEXT:    [[TMP18:%.*]] = select <4 x i1> [[EXTRACT]], <4 x i32> [[TMP11]], <4 x i32> [[X0]]
-; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
-; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <4 x i32> [[X4]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP39:%.*]] = bitcast <4 x i32> [[TMP3]] to <8 x i16>
+; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP40:%.*]] = bitcast <4 x i32> [[TMP5]] to <8 x i16>
+; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <4 x i32> [[X4]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP41:%.*]] = icmp ne <8 x i16> [[TMP39]], zeroinitializer
 ; CHECK-NEXT:    [[TMP42:%.*]] = icmp ne <8 x i16> [[TMP40]], zeroinitializer
 ; CHECK-NEXT:    [[TMP43:%.*]] = icmp ne <8 x i16> [[TMP37]], zeroinitializer
@@ -746,7 +746,7 @@ define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssd_128(<4 x i32>
 ; CHECK-NEXT:    [[TMP52:%.*]] = icmp ne <4 x i32> [[TMP51]], zeroinitializer
 ; CHECK-NEXT:    [[TMP53:%.*]] = sext <4 x i1> [[TMP52]] to <4 x i32>
 ; CHECK-NEXT:    [[_MSPROP5:%.*]] = or <4 x i32> [[TMP53]], [[TMP2]]
-; CHECK-NEXT:    [[TMP19:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X4]])
+; CHECK-NEXT:    [[TMP19:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> [[X0]], <8 x i16> [[TMP37]], <8 x i16> [[TMP38]])
 ; CHECK-NEXT:    [[TMP20:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP21:%.*]] = bitcast i8 [[X3]] to <8 x i1>
 ; CHECK-NEXT:    [[_MSPROP6:%.*]] = shufflevector <8 x i1> [[TMP20]], <8 x i1> [[TMP20]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
@@ -783,10 +783,10 @@ define <8 x i32>@test_int_x86_avx512_vpdpwssds_256(<8 x i32> %x0, <8 x i32> %x1,
 ; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
 ; CHECK-NEXT:    [[TMP21:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
-; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <8 x i32> [[TMP2]] to <16 x i16>
+; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <8 x i32> [[TMP3]] to <16 x i16>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne <16 x i16> [[TMP6]], zeroinitializer
 ; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <16 x i16> [[TMP7]], zeroinitializer
 ; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne <16 x i16> [[TMP22]], zeroinitializer
@@ -801,7 +801,7 @@ define <8 x i32>@test_int_x86_avx512_vpdpwssds_256(<8 x i32> %x0, <8 x i32> %x1,
 ; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <8 x i32> [[TMP18]], zeroinitializer
 ; CHECK-NEXT:    [[TMP20:%.*]] = sext <8 x i1> [[TMP19]] to <8 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[TMP20]], [[TMP21]]
-; CHECK-NEXT:    [[TMP4:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP4:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> [[X0]], <16 x i16> [[TMP22]], <16 x i16> [[TMP5]])
 ; CHECK-NEXT:    store <8 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <8 x i32> [[TMP4]]
 ;
@@ -829,10 +829,10 @@ define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssds_256(<8 x i32
 ; CHECK-NEXT:    [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080
 ; CHECK-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load <8 x i32>, ptr [[TMP10]], align 32
-; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
-; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP31:%.*]] = bitcast <8 x i32> [[TMP3]] to <16 x i16>
+; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP32:%.*]] = bitcast <8 x i32> [[_MSLD]] to <16 x i16>
+; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP33:%.*]] = icmp ne <16 x i16> [[TMP31]], zeroinitializer
 ; CHECK-NEXT:    [[TMP34:%.*]] = icmp ne <16 x i16> [[TMP32]], zeroinitializer
 ; CHECK-NEXT:    [[TMP35:%.*]] = icmp ne <16 x i16> [[TMP29]], zeroinitializer
@@ -847,7 +847,7 @@ define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssds_256(<8 x i32
 ; CHECK-NEXT:    [[TMP61:%.*]] = icmp ne <8 x i32> [[TMP60]], zeroinitializer
 ; CHECK-NEXT:    [[TMP62:%.*]] = sext <8 x i1> [[TMP61]] to <8 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[TMP62]], [[TMP2]]
-; CHECK-NEXT:    [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> [[X0]], <16 x i16> [[TMP29]], <16 x i16> [[TMP30]])
 ; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i8 [[X3]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP14:%.*]] = select <8 x i1> [[TMP13]], <8 x i32> [[_MSPROP1]], <8 x i32> [[TMP2]]
@@ -856,10 +856,10 @@ define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssds_256(<8 x i32
 ; CHECK-NEXT:    [[TMP17:%.*]] = or <8 x i32> [[TMP16]], [[TMP2]]
 ; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <8 x i1> [[TMP12]], <8 x i32> [[TMP17]], <8 x i32> [[TMP14]]
 ; CHECK-NEXT:    [[TMP18:%.*]] = select <8 x i1> [[TMP13]], <8 x i32> [[TMP11]], <8 x i32> [[X0]]
-; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
-; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <8 x i32> [[X4]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP39:%.*]] = bitcast <8 x i32> [[TMP3]] to <16 x i16>
+; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP40:%.*]] = bitcast <8 x i32> [[TMP5]] to <16 x i16>
+; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <8 x i32> [[X4]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP41:%.*]] = icmp ne <16 x i16> [[TMP39]], zeroinitializer
 ; CHECK-NEXT:    [[TMP42:%.*]] = icmp ne <16 x i16> [[TMP40]], zeroinitializer
 ; CHECK-NEXT:    [[TMP43:%.*]] = icmp ne <16 x i16> [[TMP37]], zeroinitializer
@@ -874,7 +874,7 @@ define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssds_256(<8 x i32
 ; CHECK-NEXT:    [[TMP52:%.*]] = icmp ne <8 x i32> [[TMP51]], zeroinitializer
 ; CHECK-NEXT:    [[TMP53:%.*]] = sext <8 x i1> [[TMP52]] to <8 x i32>
 ; CHECK-NEXT:    [[_MSPROP3:%.*]] = or <8 x i32> [[TMP53]], [[TMP2]]
-; CHECK-NEXT:    [[TMP19:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X4]])
+; CHECK-NEXT:    [[TMP19:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> [[X0]], <16 x i16> [[TMP37]], <16 x i16> [[TMP38]])
 ; CHECK-NEXT:    [[TMP20:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP21:%.*]] = bitcast i8 [[X3]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP22:%.*]] = select <8 x i1> [[TMP21]], <8 x i32> [[_MSPROP3]], <8 x i32> zeroinitializer
@@ -908,10 +908,10 @@ define <4 x i32>@test_int_x86_avx512_vpdpwssds_128(<4 x i32> %x0, <4 x i32> %x1,
 ; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
 ; CHECK-NEXT:    [[TMP21:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
-; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <4 x i32> [[TMP2]] to <8 x i16>
+; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <4 x i32> [[TMP3]] to <8 x i16>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne <8 x i16> [[TMP6]], zeroinitializer
 ; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <8 x i16> [[TMP7]], zeroinitializer
 ; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne <8 x i16> [[TMP22]], zeroinitializer
@@ -926,7 +926,7 @@ define <4 x i32>@test_int_x86_avx512_vpdpwssds_128(<4 x i32> %x0, <4 x i32> %x1,
 ; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <4 x i32> [[TMP18]], zeroinitializer
 ; CHECK-NEXT:    [[TMP20:%.*]] = sext <4 x i1> [[TMP19]] to <4 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[TMP20]], [[TMP21]]
-; CHECK-NEXT:    [[TMP4:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP4:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> [[X0]], <8 x i16> [[TMP22]], <8 x i16> [[TMP5]])
 ; CHECK-NEXT:    store <4 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <4 x i32> [[TMP4]]
 ;
@@ -954,10 +954,10 @@ define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssds_128(<4 x i32
 ; CHECK-NEXT:    [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080
 ; CHECK-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load <4 x i32>, ptr [[TMP10]], align 16
-; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
-; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP31:%.*]] = bitcast <4 x i32> [[TMP3]] to <8 x i16>
+; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP32:%.*]] = bitcast <4 x i32> [[_MSLD]] to <8 x i16>
+; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP33:%.*]] = icmp ne <8 x i16> [[TMP31]], zeroinitializer
 ; CHECK-NEXT:    [[TMP34:%.*]] = icmp ne <8 x i16> [[TMP32]], zeroinitializer
 ; CHECK-NEXT:    [[TMP35:%.*]] = icmp ne <8 x i16> [[TMP29]], zeroinitializer
@@ -972,7 +972,7 @@ define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssds_128(<4 x i32
 ; CHECK-NEXT:    [[TMP61:%.*]] = icmp ne <4 x i32> [[TMP60]], zeroinitializer
 ; CHECK-NEXT:    [[TMP62:%.*]] = sext <4 x i1> [[TMP61]] to <4 x i32>
 ; CHECK-NEXT:    [[_MSPROP2:%.*]] = or <4 x i32> [[TMP62]], [[TMP2]]
-; CHECK-NEXT:    [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> [[X0]], <8 x i16> [[TMP29]], <8 x i16> [[TMP30]])
 ; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i8 [[X3]] to <8 x i1>
 ; CHECK-NEXT:    [[_MSPROP3:%.*]] = shufflevector <8 x i1> [[TMP12]], <8 x i1> [[TMP12]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
@@ -983,10 +983,10 @@ define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssds_128(<4 x i32
 ; CHECK-NEXT:    [[TMP17:%.*]] = or <4 x i32> [[TMP16]], [[TMP2]]
 ; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <4 x i1> [[_MSPROP3]], <4 x i32> [[TMP17]], <4 x i32> [[TMP14]]
 ; CHECK-NEXT:    [[TMP18:%.*]] = select <4 x i1> [[EXTRACT]], <4 x i32> [[TMP11]], <4 x i32> [[X0]]
-; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
-; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <4 x i32> [[X4]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP39:%.*]] = bitcast <4 x i32> [[TMP3]] to <8 x i16>
+; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP40:%.*]] = bitcast <4 x i32> [[TMP5]] to <8 x i16>
+; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <4 x i32> [[X4]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP41:%.*]] = icmp ne <8 x i16> [[TMP39]], zeroinitializer
 ; CHECK-NEXT:    [[TMP42:%.*]] = icmp ne <8 x i16> [[TMP40]], zeroinitializer
 ; CHECK-NEXT:    [[TMP43:%.*]] = icmp ne <8 x i16> [[TMP37]], zeroinitializer
@@ -1001,7 +1001,7 @@ define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssds_128(<4 x i32
 ; CHECK-NEXT:    [[TMP52:%.*]] = icmp ne <4 x i32> [[TMP51]], zeroinitializer
 ; CHECK-NEXT:    [[TMP53:%.*]] = sext <4 x i1> [[TMP52]] to <4 x i32>
 ; CHECK-NEXT:    [[_MSPROP5:%.*]] = or <4 x i32> [[TMP53]], [[TMP2]]
-; CHECK-NEXT:    [[TMP19:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X4]])
+; CHECK-NEXT:    [[TMP19:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> [[X0]], <8 x i16> [[TMP37]], <8 x i16> [[TMP38]])
 ; CHECK-NEXT:    [[TMP20:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP21:%.*]] = bitcast i8 [[X3]] to <8 x i1>
 ; CHECK-NEXT:    [[_MSPROP6:%.*]] = shufflevector <8 x i1> [[TMP20]], <8 x i1> [[TMP20]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
diff --git a/llvm/test/Instrumentation/MemorySanitizer/X86/avx512vl_vnni-intrinsics.ll b/llvm/test/Instrumentation/MemorySanitizer/X86/avx512vl_vnni-intrinsics.ll
index 1d3046804b74f..f2b1e16e3d5c2 100644
--- a/llvm/test/Instrumentation/MemorySanitizer/X86/avx512vl_vnni-intrinsics.ll
+++ b/llvm/test/Instrumentation/MemorySanitizer/X86/avx512vl_vnni-intrinsics.ll
@@ -495,10 +495,10 @@ define <8 x i32>@test_int_x86_avx512_vpdpwssd_256(<8 x i32> %x0, <8 x i32> %x1,
 ; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
 ; CHECK-NEXT:    [[TMP21:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
-; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <8 x i32> [[TMP2]] to <16 x i16>
+; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <8 x i32> [[TMP3]] to <16 x i16>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne <16 x i16> [[TMP6]], zeroinitializer
 ; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <16 x i16> [[TMP7]], zeroinitializer
 ; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne <16 x i16> [[TMP22]], zeroinitializer
@@ -513,7 +513,7 @@ define <8 x i32>@test_int_x86_avx512_vpdpwssd_256(<8 x i32> %x0, <8 x i32> %x1,
 ; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <8 x i32> [[TMP18]], zeroinitializer
 ; CHECK-NEXT:    [[TMP20:%.*]] = sext <8 x i1> [[TMP19]] to <8 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[TMP20]], [[TMP21]]
-; CHECK-NEXT:    [[TMP4:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP4:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> [[X0]], <16 x i16> [[TMP22]], <16 x i16> [[TMP5]])
 ; CHECK-NEXT:    store <8 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <8 x i32> [[TMP4]]
 ;
@@ -541,10 +541,10 @@ define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssd_256(<8 x i32>
 ; CHECK-NEXT:    [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080
 ; CHECK-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load <8 x i32>, ptr [[TMP10]], align 32
-; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
-; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP31:%.*]] = bitcast <8 x i32> [[TMP3]] to <16 x i16>
+; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP32:%.*]] = bitcast <8 x i32> [[_MSLD]] to <16 x i16>
+; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP33:%.*]] = icmp ne <16 x i16> [[TMP31]], zeroinitializer
 ; CHECK-NEXT:    [[TMP34:%.*]] = icmp ne <16 x i16> [[TMP32]], zeroinitializer
 ; CHECK-NEXT:    [[TMP35:%.*]] = icmp ne <16 x i16> [[TMP29]], zeroinitializer
@@ -559,7 +559,7 @@ define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssd_256(<8 x i32>
 ; CHECK-NEXT:    [[TMP61:%.*]] = icmp ne <8 x i32> [[TMP60]], zeroinitializer
 ; CHECK-NEXT:    [[TMP62:%.*]] = sext <8 x i1> [[TMP61]] to <8 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[TMP62]], [[TMP2]]
-; CHECK-NEXT:    [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> [[X0]], <16 x i16> [[TMP29]], <16 x i16> [[TMP30]])
 ; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i8 [[X3]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP14:%.*]] = select <8 x i1> [[TMP13]], <8 x i32> [[_MSPROP1]], <8 x i32> [[TMP2]]
@@ -568,10 +568,10 @@ define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssd_256(<8 x i32>
 ; CHECK-NEXT:    [[TMP17:%.*]] = or <8 x i32> [[TMP16]], [[TMP2]]
 ; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <8 x i1> [[TMP12]], <8 x i32> [[TMP17]], <8 x i32> [[TMP14]]
 ; CHECK-NEXT:    [[TMP18:%.*]] = select <8 x i1> [[TMP13]], <8 x i32> [[TMP11]], <8 x i32> [[X0]]
-; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
-; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <8 x i32> [[X4]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP39:%.*]] = bitcast <8 x i32> [[TMP3]] to <16 x i16>
+; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP40:%.*]] = bitcast <8 x i32> [[TMP5]] to <16 x i16>
+; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <8 x i32> [[X4]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP41:%.*]] = icmp ne <16 x i16> [[TMP39]], zeroinitializer
 ; CHECK-NEXT:    [[TMP42:%.*]] = icmp ne <16 x i16> [[TMP40]], zeroinitializer
 ; CHECK-NEXT:    [[TMP43:%.*]] = icmp ne <16 x i16> [[TMP37]], zeroinitializer
@@ -586,7 +586,7 @@ define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssd_256(<8 x i32>
 ; CHECK-NEXT:    [[TMP52:%.*]] = icmp ne <8 x i32> [[TMP51]], zeroinitializer
 ; CHECK-NEXT:    [[TMP53:%.*]] = sext <8 x i1> [[TMP52]] to <8 x i32>
 ; CHECK-NEXT:    [[_MSPROP3:%.*]] = or <8 x i32> [[TMP53]], [[TMP2]]
-; CHECK-NEXT:    [[TMP19:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X4]])
+; CHECK-NEXT:    [[TMP19:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> [[X0]], <16 x i16> [[TMP37]], <16 x i16> [[TMP38]])
 ; CHECK-NEXT:    [[TMP20:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP21:%.*]] = bitcast i8 [[X3]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP22:%.*]] = select <8 x i1> [[TMP21]], <8 x i32> [[_MSPROP3]], <8 x i32> zeroinitializer
@@ -623,10 +623,10 @@ define <4 x i32>@test_int_x86_avx512_vpdpwssd_128(<4 x i32> %x0, <4 x i32> %x1,
 ; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
 ; CHECK-NEXT:    [[TMP21:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
-; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <4 x i32> [[TMP2]] to <8 x i16>
+; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <4 x i32> [[TMP3]] to <8 x i16>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne <8 x i16> [[TMP6]], zeroinitializer
 ; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <8 x i16> [[TMP7]], zeroinitializer
 ; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne <8 x i16> [[TMP22]], zeroinitializer
@@ -641,7 +641,7 @@ define <4 x i32>@test_int_x86_avx512_vpdpwssd_128(<4 x i32> %x0, <4 x i32> %x1,
 ; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <4 x i32> [[TMP18]], zeroinitializer
 ; CHECK-NEXT:    [[TMP20:%.*]] = sext <4 x i1> [[TMP19]] to <4 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[TMP20]], [[TMP21]]
-; CHECK-NEXT:    [[TMP4:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP4:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> [[X0]], <8 x i16> [[TMP22]], <8 x i16> [[TMP5]])
 ; CHECK-NEXT:    store <4 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <4 x i32> [[TMP4]]
 ;
@@ -669,10 +669,10 @@ define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssd_128(<4 x i32>
 ; CHECK-NEXT:    [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080
 ; CHECK-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load <4 x i32>, ptr [[TMP10]], align 16
-; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
-; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP31:%.*]] = bitcast <4 x i32> [[TMP3]] to <8 x i16>
+; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP32:%.*]] = bitcast <4 x i32> [[_MSLD]] to <8 x i16>
+; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP33:%.*]] = icmp ne <8 x i16> [[TMP31]], zeroinitializer
 ; CHECK-NEXT:    [[TMP34:%.*]] = icmp ne <8 x i16> [[TMP32]], zeroinitializer
 ; CHECK-NEXT:    [[TMP35:%.*]] = icmp ne <8 x i16> [[TMP29]], zeroinitializer
@@ -687,7 +687,7 @@ define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssd_128(<4 x i32>
 ; CHECK-NEXT:    [[TMP61:%.*]] = icmp ne <4 x i32> [[TMP60]], zeroinitializer
 ; CHECK-NEXT:    [[TMP62:%.*]] = sext <4 x i1> [[TMP61]] to <4 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[TMP62]], [[TMP2]]
-; CHECK-NEXT:    [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> [[X0]], <8 x i16> [[TMP29]], <8 x i16> [[TMP30]])
 ; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i8 [[X3]] to <8 x i1>
 ; CHECK-NEXT:    [[_MSPROP2:%.*]] = shufflevector <8 x i1> [[TMP12]], <8 x i1> [[TMP12]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
@@ -698,10 +698,10 @@ define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssd_128(<4 x i32>
 ; CHECK-NEXT:    [[TMP17:%.*]] = or <4 x i32> [[TMP16]], [[TMP2]]
 ; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <4 x i1> [[_MSPROP2]], <4 x i32> [[TMP17]], <4 x i32> [[TMP14]]
 ; CHECK-NEXT:    [[TMP18:%.*]] = select <4 x i1> [[EXTRACT]], <4 x i32> [[TMP11]], <4 x i32> [[X0]]
-; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
-; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <4 x i32> [[X4]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP39:%.*]] = bitcast <4 x i32> [[TMP3]] to <8 x i16>
+; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP40:%.*]] = bitcast <4 x i32> [[TMP5]] to <8 x i16>
+; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <4 x i32> [[X4]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP41:%.*]] = icmp ne <8 x i16> [[TMP39]], zeroinitializer
 ; CHECK-NEXT:    [[TMP42:%.*]] = icmp ne <8 x i16> [[TMP40]], zeroinitializer
 ; CHECK-NEXT:    [[TMP43:%.*]] = icmp ne <8 x i16> [[TMP37]], zeroinitializer
@@ -716,7 +716,7 @@ define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssd_128(<4 x i32>
 ; CHECK-NEXT:    [[TMP52:%.*]] = icmp ne <4 x i32> [[TMP51]], zeroinitializer
 ; CHECK-NEXT:    [[TMP53:%.*]] = sext <4 x i1> [[TMP52]] to <4 x i32>
 ; CHECK-NEXT:    [[_MSPROP4:%.*]] = or <4 x i32> [[TMP53]], [[TMP2]]
-; CHECK-NEXT:    [[TMP19:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X4]])
+; CHECK-NEXT:    [[TMP19:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> [[X0]], <8 x i16> [[TMP37]], <8 x i16> [[TMP38]])
 ; CHECK-NEXT:    [[TMP20:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP21:%.*]] = bitcast i8 [[X3]] to <8 x i1>
 ; CHECK-NEXT:    [[_MSPROP5:%.*]] = shufflevector <8 x i1> [[TMP20]], <8 x i1> [[TMP20]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
@@ -757,10 +757,10 @@ define <8 x i32>@test_int_x86_avx512_vpdpwssds_256(<8 x i32> %x0, <8 x i32> %x1,
 ; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
 ; CHECK-NEXT:    [[TMP21:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
-; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <8 x i32> [[TMP2]] to <16 x i16>
+; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <8 x i32> [[TMP3]] to <16 x i16>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne <16 x i16> [[TMP6]], zeroinitializer
 ; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <16 x i16> [[TMP7]], zeroinitializer
 ; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne <16 x i16> [[TMP22]], zeroinitializer
@@ -775,7 +775,7 @@ define <8 x i32>@test_int_x86_avx512_vpdpwssds_256(<8 x i32> %x0, <8 x i32> %x1,
 ; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <8 x i32> [[TMP18]], zeroinitializer
 ; CHECK-NEXT:    [[TMP20:%.*]] = sext <8 x i1> [[TMP19]] to <8 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[TMP20]], [[TMP21]]
-; CHECK-NEXT:    [[TMP4:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP4:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> [[X0]], <16 x i16> [[TMP22]], <16 x i16> [[TMP5]])
 ; CHECK-NEXT:    store <8 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <8 x i32> [[TMP4]]
 ;
@@ -803,10 +803,10 @@ define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssds_256(<8 x i32
 ; CHECK-NEXT:    [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080
 ; CHECK-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load <8 x i32>, ptr [[TMP10]], align 32
-; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
-; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP31:%.*]] = bitcast <8 x i32> [[TMP3]] to <16 x i16>
+; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP32:%.*]] = bitcast <8 x i32> [[_MSLD]] to <16 x i16>
+; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP33:%.*]] = icmp ne <16 x i16> [[TMP31]], zeroinitializer
 ; CHECK-NEXT:    [[TMP34:%.*]] = icmp ne <16 x i16> [[TMP32]], zeroinitializer
 ; CHECK-NEXT:    [[TMP35:%.*]] = icmp ne <16 x i16> [[TMP29]], zeroinitializer
@@ -821,7 +821,7 @@ define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssds_256(<8 x i32
 ; CHECK-NEXT:    [[TMP61:%.*]] = icmp ne <8 x i32> [[TMP60]], zeroinitializer
 ; CHECK-NEXT:    [[TMP62:%.*]] = sext <8 x i1> [[TMP61]] to <8 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[TMP62]], [[TMP2]]
-; CHECK-NEXT:    [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> [[X0]], <16 x i16> [[TMP29]], <16 x i16> [[TMP30]])
 ; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i8 [[X3]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP14:%.*]] = select <8 x i1> [[TMP13]], <8 x i32> [[_MSPROP1]], <8 x i32> [[TMP2]]
@@ -830,10 +830,10 @@ define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssds_256(<8 x i32
 ; CHECK-NEXT:    [[TMP17:%.*]] = or <8 x i32> [[TMP16]], [[TMP2]]
 ; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <8 x i1> [[TMP12]], <8 x i32> [[TMP17]], <8 x i32> [[TMP14]]
 ; CHECK-NEXT:    [[TMP18:%.*]] = select <8 x i1> [[TMP13]], <8 x i32> [[TMP11]], <8 x i32> [[X0]]
-; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
-; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <8 x i32> [[X4]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP39:%.*]] = bitcast <8 x i32> [[TMP3]] to <16 x i16>
+; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP40:%.*]] = bitcast <8 x i32> [[TMP5]] to <16 x i16>
+; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <8 x i32> [[X4]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP41:%.*]] = icmp ne <16 x i16> [[TMP39]], zeroinitializer
 ; CHECK-NEXT:    [[TMP42:%.*]] = icmp ne <16 x i16> [[TMP40]], zeroinitializer
 ; CHECK-NEXT:    [[TMP43:%.*]] = icmp ne <16 x i16> [[TMP37]], zeroinitializer
@@ -848,7 +848,7 @@ define { <8 x i32>, <8 x i32> } @test_int_x86_avx512_mask_vpdpwssds_256(<8 x i32
 ; CHECK-NEXT:    [[TMP52:%.*]] = icmp ne <8 x i32> [[TMP51]], zeroinitializer
 ; CHECK-NEXT:    [[TMP53:%.*]] = sext <8 x i1> [[TMP52]] to <8 x i32>
 ; CHECK-NEXT:    [[_MSPROP3:%.*]] = or <8 x i32> [[TMP53]], [[TMP2]]
-; CHECK-NEXT:    [[TMP19:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X4]])
+; CHECK-NEXT:    [[TMP19:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> [[X0]], <16 x i16> [[TMP37]], <16 x i16> [[TMP38]])
 ; CHECK-NEXT:    [[TMP20:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP21:%.*]] = bitcast i8 [[X3]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP22:%.*]] = select <8 x i1> [[TMP21]], <8 x i32> [[_MSPROP3]], <8 x i32> zeroinitializer
@@ -896,10 +896,10 @@ define <4 x i32>@test_int_x86_avx512_vpdpwssds_128(<4 x i32> %x0, <4 x i32> %x1,
 ; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
 ; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load <4 x i32>, ptr [[TMP8]], align 16
-; CHECK-NEXT:    [[TMP26:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
-; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP11:%.*]] = bitcast <4 x i32> [[TMP3]] to <8 x i16>
+; CHECK-NEXT:    [[TMP26:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP12:%.*]] = bitcast <4 x i32> [[_MSLD]] to <8 x i16>
+; CHECK-NEXT:    [[TMP10:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP13:%.*]] = icmp ne <8 x i16> [[TMP11]], zeroinitializer
 ; CHECK-NEXT:    [[TMP14:%.*]] = icmp ne <8 x i16> [[TMP12]], zeroinitializer
 ; CHECK-NEXT:    [[TMP15:%.*]] = icmp ne <8 x i16> [[TMP26]], zeroinitializer
@@ -914,7 +914,7 @@ define <4 x i32>@test_int_x86_avx512_vpdpwssds_128(<4 x i32> %x0, <4 x i32> %x1,
 ; CHECK-NEXT:    [[TMP24:%.*]] = icmp ne <4 x i32> [[TMP23]], zeroinitializer
 ; CHECK-NEXT:    [[TMP25:%.*]] = sext <4 x i1> [[TMP24]] to <4 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[TMP25]], [[TMP4]]
-; CHECK-NEXT:    [[TMP9:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP9:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> [[X0]], <8 x i16> [[TMP26]], <8 x i16> [[TMP10]])
 ; CHECK-NEXT:    store <4 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <4 x i32> [[TMP9]]
 ;
@@ -943,10 +943,10 @@ define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssds_128(<4 x i32
 ; CHECK-NEXT:    [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080
 ; CHECK-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load <4 x i32>, ptr [[TMP10]], align 16
-; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
-; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP31:%.*]] = bitcast <4 x i32> [[TMP3]] to <8 x i16>
+; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP32:%.*]] = bitcast <4 x i32> [[_MSLD]] to <8 x i16>
+; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP33:%.*]] = icmp ne <8 x i16> [[TMP31]], zeroinitializer
 ; CHECK-NEXT:    [[TMP34:%.*]] = icmp ne <8 x i16> [[TMP32]], zeroinitializer
 ; CHECK-NEXT:    [[TMP35:%.*]] = icmp ne <8 x i16> [[TMP29]], zeroinitializer
@@ -961,7 +961,7 @@ define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssds_128(<4 x i32
 ; CHECK-NEXT:    [[TMP61:%.*]] = icmp ne <4 x i32> [[TMP60]], zeroinitializer
 ; CHECK-NEXT:    [[TMP62:%.*]] = sext <4 x i1> [[TMP61]] to <4 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[TMP62]], [[TMP2]]
-; CHECK-NEXT:    [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> [[X0]], <8 x i16> [[TMP29]], <8 x i16> [[TMP30]])
 ; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i8 [[X3]] to <8 x i1>
 ; CHECK-NEXT:    [[_MSPROP2:%.*]] = shufflevector <8 x i1> [[TMP12]], <8 x i1> [[TMP12]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
@@ -972,10 +972,10 @@ define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssds_128(<4 x i32
 ; CHECK-NEXT:    [[TMP17:%.*]] = or <4 x i32> [[TMP16]], [[TMP2]]
 ; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <4 x i1> [[_MSPROP2]], <4 x i32> [[TMP17]], <4 x i32> [[TMP14]]
 ; CHECK-NEXT:    [[TMP18:%.*]] = select <4 x i1> [[EXTRACT]], <4 x i32> [[TMP11]], <4 x i32> [[X0]]
-; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
-; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <4 x i32> [[X4]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP39:%.*]] = bitcast <4 x i32> [[TMP3]] to <8 x i16>
+; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP40:%.*]] = bitcast <4 x i32> [[TMP5]] to <8 x i16>
+; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <4 x i32> [[X4]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP41:%.*]] = icmp ne <8 x i16> [[TMP39]], zeroinitializer
 ; CHECK-NEXT:    [[TMP42:%.*]] = icmp ne <8 x i16> [[TMP40]], zeroinitializer
 ; CHECK-NEXT:    [[TMP43:%.*]] = icmp ne <8 x i16> [[TMP37]], zeroinitializer
@@ -990,7 +990,7 @@ define { <4 x i32>, <4 x i32> } @test_int_x86_avx512_mask_vpdpwssds_128(<4 x i32
 ; CHECK-NEXT:    [[TMP52:%.*]] = icmp ne <4 x i32> [[TMP51]], zeroinitializer
 ; CHECK-NEXT:    [[TMP53:%.*]] = sext <4 x i1> [[TMP52]] to <4 x i32>
 ; CHECK-NEXT:    [[_MSPROP4:%.*]] = or <4 x i32> [[TMP53]], [[TMP2]]
-; CHECK-NEXT:    [[TMP19:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X4]])
+; CHECK-NEXT:    [[TMP19:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> [[X0]], <8 x i16> [[TMP37]], <8 x i16> [[TMP38]])
 ; CHECK-NEXT:    [[TMP20:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP21:%.*]] = bitcast i8 [[X3]] to <8 x i1>
 ; CHECK-NEXT:    [[_MSPROP5:%.*]] = shufflevector <8 x i1> [[TMP20]], <8 x i1> [[TMP20]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
diff --git a/llvm/test/Instrumentation/MemorySanitizer/X86/avx512vnni-intrinsics-upgrade.ll b/llvm/test/Instrumentation/MemorySanitizer/X86/avx512vnni-intrinsics-upgrade.ll
index 5c99f8a3a1fb6..4e7598b92abcf 100644
--- a/llvm/test/Instrumentation/MemorySanitizer/X86/avx512vnni-intrinsics-upgrade.ll
+++ b/llvm/test/Instrumentation/MemorySanitizer/X86/avx512vnni-intrinsics-upgrade.ll
@@ -270,10 +270,10 @@ define <16 x i32>@test_int_x86_avx512_vpdpwssd_512(<16 x i32> %x0, <16 x i32> %x
 ; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 128), align 8
 ; CHECK-NEXT:    [[TMP21:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
-; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <16 x i32> [[X2]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <16 x i32> [[TMP2]] to <32 x i16>
+; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <16 x i32> [[TMP3]] to <32 x i16>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <16 x i32> [[X2]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne <32 x i16> [[TMP6]], zeroinitializer
 ; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <32 x i16> [[TMP7]], zeroinitializer
 ; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne <32 x i16> [[TMP22]], zeroinitializer
@@ -288,7 +288,7 @@ define <16 x i32>@test_int_x86_avx512_vpdpwssd_512(<16 x i32> %x0, <16 x i32> %x
 ; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <16 x i32> [[TMP18]], zeroinitializer
 ; CHECK-NEXT:    [[TMP20:%.*]] = sext <16 x i1> [[TMP19]] to <16 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[TMP20]], [[TMP21]]
-; CHECK-NEXT:    [[TMP4:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> [[X0]], <16 x i32> [[X1]], <16 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP4:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> [[X0]], <32 x i16> [[TMP22]], <32 x i16> [[TMP5]])
 ; CHECK-NEXT:    store <16 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <16 x i32> [[TMP4]]
 ;
@@ -316,10 +316,10 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssd_512(<16 x i
 ; CHECK-NEXT:    [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080
 ; CHECK-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP10]], align 64
-; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
-; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <16 x i32> [[X2]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP31:%.*]] = bitcast <16 x i32> [[TMP3]] to <32 x i16>
+; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP32:%.*]] = bitcast <16 x i32> [[_MSLD]] to <32 x i16>
+; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <16 x i32> [[X2]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP33:%.*]] = icmp ne <32 x i16> [[TMP31]], zeroinitializer
 ; CHECK-NEXT:    [[TMP34:%.*]] = icmp ne <32 x i16> [[TMP32]], zeroinitializer
 ; CHECK-NEXT:    [[TMP35:%.*]] = icmp ne <32 x i16> [[TMP29]], zeroinitializer
@@ -334,7 +334,7 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssd_512(<16 x i
 ; CHECK-NEXT:    [[TMP61:%.*]] = icmp ne <16 x i32> [[TMP60]], zeroinitializer
 ; CHECK-NEXT:    [[TMP62:%.*]] = sext <16 x i1> [[TMP61]] to <16 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[TMP62]], [[TMP2]]
-; CHECK-NEXT:    [[TMP11:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> [[X0]], <16 x i32> [[X1]], <16 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP11:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> [[X0]], <32 x i16> [[TMP29]], <32 x i16> [[TMP30]])
 ; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i16 [[TMP4]] to <16 x i1>
 ; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i16 [[X3]] to <16 x i1>
 ; CHECK-NEXT:    [[TMP14:%.*]] = select <16 x i1> [[TMP13]], <16 x i32> [[_MSPROP1]], <16 x i32> [[TMP2]]
@@ -343,10 +343,10 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssd_512(<16 x i
 ; CHECK-NEXT:    [[TMP17:%.*]] = or <16 x i32> [[TMP16]], [[TMP2]]
 ; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <16 x i1> [[TMP12]], <16 x i32> [[TMP17]], <16 x i32> [[TMP14]]
 ; CHECK-NEXT:    [[TMP18:%.*]] = select <16 x i1> [[TMP13]], <16 x i32> [[TMP11]], <16 x i32> [[X0]]
-; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
-; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <16 x i32> [[X4]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP39:%.*]] = bitcast <16 x i32> [[TMP3]] to <32 x i16>
+; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP40:%.*]] = bitcast <16 x i32> [[TMP5]] to <32 x i16>
+; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <16 x i32> [[X4]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP41:%.*]] = icmp ne <32 x i16> [[TMP39]], zeroinitializer
 ; CHECK-NEXT:    [[TMP42:%.*]] = icmp ne <32 x i16> [[TMP40]], zeroinitializer
 ; CHECK-NEXT:    [[TMP43:%.*]] = icmp ne <32 x i16> [[TMP37]], zeroinitializer
@@ -361,7 +361,7 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssd_512(<16 x i
 ; CHECK-NEXT:    [[TMP52:%.*]] = icmp ne <16 x i32> [[TMP51]], zeroinitializer
 ; CHECK-NEXT:    [[TMP53:%.*]] = sext <16 x i1> [[TMP52]] to <16 x i32>
 ; CHECK-NEXT:    [[_MSPROP3:%.*]] = or <16 x i32> [[TMP53]], [[TMP2]]
-; CHECK-NEXT:    [[TMP19:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> [[X0]], <16 x i32> [[X1]], <16 x i32> [[X4]])
+; CHECK-NEXT:    [[TMP19:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> [[X0]], <32 x i16> [[TMP37]], <32 x i16> [[TMP38]])
 ; CHECK-NEXT:    [[TMP20:%.*]] = bitcast i16 [[TMP4]] to <16 x i1>
 ; CHECK-NEXT:    [[TMP21:%.*]] = bitcast i16 [[X3]] to <16 x i1>
 ; CHECK-NEXT:    [[TMP22:%.*]] = select <16 x i1> [[TMP21]], <16 x i32> [[_MSPROP3]], <16 x i32> zeroinitializer
@@ -395,10 +395,10 @@ define <16 x i32>@test_int_x86_avx512_vpdpwssds_512(<16 x i32> %x0, <16 x i32> %
 ; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 128), align 8
 ; CHECK-NEXT:    [[TMP21:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
-; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <16 x i32> [[X2]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <16 x i32> [[TMP2]] to <32 x i16>
+; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <16 x i32> [[TMP3]] to <32 x i16>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <16 x i32> [[X2]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne <32 x i16> [[TMP6]], zeroinitializer
 ; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <32 x i16> [[TMP7]], zeroinitializer
 ; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne <32 x i16> [[TMP22]], zeroinitializer
@@ -413,7 +413,7 @@ define <16 x i32>@test_int_x86_avx512_vpdpwssds_512(<16 x i32> %x0, <16 x i32> %
 ; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <16 x i32> [[TMP18]], zeroinitializer
 ; CHECK-NEXT:    [[TMP20:%.*]] = sext <16 x i1> [[TMP19]] to <16 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[TMP20]], [[TMP21]]
-; CHECK-NEXT:    [[TMP4:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> [[X0]], <16 x i32> [[X1]], <16 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP4:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> [[X0]], <32 x i16> [[TMP22]], <32 x i16> [[TMP5]])
 ; CHECK-NEXT:    store <16 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <16 x i32> [[TMP4]]
 ;
@@ -441,10 +441,10 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssds_512(<16 x
 ; CHECK-NEXT:    [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080
 ; CHECK-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP10]], align 64
-; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
-; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <16 x i32> [[X2]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP31:%.*]] = bitcast <16 x i32> [[TMP3]] to <32 x i16>
+; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP32:%.*]] = bitcast <16 x i32> [[_MSLD]] to <32 x i16>
+; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <16 x i32> [[X2]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP33:%.*]] = icmp ne <32 x i16> [[TMP31]], zeroinitializer
 ; CHECK-NEXT:    [[TMP34:%.*]] = icmp ne <32 x i16> [[TMP32]], zeroinitializer
 ; CHECK-NEXT:    [[TMP35:%.*]] = icmp ne <32 x i16> [[TMP29]], zeroinitializer
@@ -459,7 +459,7 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssds_512(<16 x
 ; CHECK-NEXT:    [[TMP61:%.*]] = icmp ne <16 x i32> [[TMP60]], zeroinitializer
 ; CHECK-NEXT:    [[TMP62:%.*]] = sext <16 x i1> [[TMP61]] to <16 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[TMP62]], [[TMP2]]
-; CHECK-NEXT:    [[TMP11:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> [[X0]], <16 x i32> [[X1]], <16 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP11:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> [[X0]], <32 x i16> [[TMP29]], <32 x i16> [[TMP30]])
 ; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i16 [[TMP4]] to <16 x i1>
 ; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i16 [[X3]] to <16 x i1>
 ; CHECK-NEXT:    [[TMP14:%.*]] = select <16 x i1> [[TMP13]], <16 x i32> [[_MSPROP1]], <16 x i32> [[TMP2]]
@@ -468,10 +468,10 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssds_512(<16 x
 ; CHECK-NEXT:    [[TMP17:%.*]] = or <16 x i32> [[TMP16]], [[TMP2]]
 ; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <16 x i1> [[TMP12]], <16 x i32> [[TMP17]], <16 x i32> [[TMP14]]
 ; CHECK-NEXT:    [[TMP18:%.*]] = select <16 x i1> [[TMP13]], <16 x i32> [[TMP11]], <16 x i32> [[X0]]
-; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
-; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <16 x i32> [[X4]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP39:%.*]] = bitcast <16 x i32> [[TMP3]] to <32 x i16>
+; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP40:%.*]] = bitcast <16 x i32> [[TMP5]] to <32 x i16>
+; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <16 x i32> [[X4]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP41:%.*]] = icmp ne <32 x i16> [[TMP39]], zeroinitializer
 ; CHECK-NEXT:    [[TMP42:%.*]] = icmp ne <32 x i16> [[TMP40]], zeroinitializer
 ; CHECK-NEXT:    [[TMP43:%.*]] = icmp ne <32 x i16> [[TMP37]], zeroinitializer
@@ -486,7 +486,7 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssds_512(<16 x
 ; CHECK-NEXT:    [[TMP52:%.*]] = icmp ne <16 x i32> [[TMP51]], zeroinitializer
 ; CHECK-NEXT:    [[TMP53:%.*]] = sext <16 x i1> [[TMP52]] to <16 x i32>
 ; CHECK-NEXT:    [[_MSPROP3:%.*]] = or <16 x i32> [[TMP53]], [[TMP2]]
-; CHECK-NEXT:    [[TMP19:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> [[X0]], <16 x i32> [[X1]], <16 x i32> [[X4]])
+; CHECK-NEXT:    [[TMP19:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> [[X0]], <32 x i16> [[TMP37]], <32 x i16> [[TMP38]])
 ; CHECK-NEXT:    [[TMP20:%.*]] = bitcast i16 [[TMP4]] to <16 x i1>
 ; CHECK-NEXT:    [[TMP21:%.*]] = bitcast i16 [[X3]] to <16 x i1>
 ; CHECK-NEXT:    [[TMP22:%.*]] = select <16 x i1> [[TMP21]], <16 x i32> [[_MSPROP3]], <16 x i32> zeroinitializer
diff --git a/llvm/test/Instrumentation/MemorySanitizer/X86/avx512vnni-intrinsics.ll b/llvm/test/Instrumentation/MemorySanitizer/X86/avx512vnni-intrinsics.ll
index 236ff45c6cd08..0fd60149cc15e 100644
--- a/llvm/test/Instrumentation/MemorySanitizer/X86/avx512vnni-intrinsics.ll
+++ b/llvm/test/Instrumentation/MemorySanitizer/X86/avx512vnni-intrinsics.ll
@@ -251,10 +251,10 @@ define <16 x i32>@test_int_x86_avx512_vpdpwssd_512(<16 x i32> %x0, <16 x i32> %x
 ; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 128), align 8
 ; CHECK-NEXT:    [[TMP21:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
-; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <16 x i32> [[X2]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <16 x i32> [[TMP2]] to <32 x i16>
+; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <16 x i32> [[TMP3]] to <32 x i16>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <16 x i32> [[X2]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne <32 x i16> [[TMP6]], zeroinitializer
 ; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <32 x i16> [[TMP7]], zeroinitializer
 ; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne <32 x i16> [[TMP22]], zeroinitializer
@@ -269,7 +269,7 @@ define <16 x i32>@test_int_x86_avx512_vpdpwssd_512(<16 x i32> %x0, <16 x i32> %x
 ; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <16 x i32> [[TMP18]], zeroinitializer
 ; CHECK-NEXT:    [[TMP20:%.*]] = sext <16 x i1> [[TMP19]] to <16 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[TMP20]], [[TMP21]]
-; CHECK-NEXT:    [[TMP4:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> [[X0]], <16 x i32> [[X1]], <16 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP4:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> [[X0]], <32 x i16> [[TMP22]], <32 x i16> [[TMP5]])
 ; CHECK-NEXT:    store <16 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <16 x i32> [[TMP4]]
 ;
@@ -297,10 +297,10 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssd_512(<16 x i
 ; CHECK-NEXT:    [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080
 ; CHECK-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP10]], align 64
-; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
-; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <16 x i32> [[X2]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP31:%.*]] = bitcast <16 x i32> [[TMP3]] to <32 x i16>
+; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP32:%.*]] = bitcast <16 x i32> [[_MSLD]] to <32 x i16>
+; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <16 x i32> [[X2]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP33:%.*]] = icmp ne <32 x i16> [[TMP31]], zeroinitializer
 ; CHECK-NEXT:    [[TMP34:%.*]] = icmp ne <32 x i16> [[TMP32]], zeroinitializer
 ; CHECK-NEXT:    [[TMP35:%.*]] = icmp ne <32 x i16> [[TMP29]], zeroinitializer
@@ -315,7 +315,7 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssd_512(<16 x i
 ; CHECK-NEXT:    [[TMP61:%.*]] = icmp ne <16 x i32> [[TMP60]], zeroinitializer
 ; CHECK-NEXT:    [[TMP62:%.*]] = sext <16 x i1> [[TMP61]] to <16 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[TMP62]], [[TMP2]]
-; CHECK-NEXT:    [[TMP11:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> [[X0]], <16 x i32> [[X1]], <16 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP11:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> [[X0]], <32 x i16> [[TMP29]], <32 x i16> [[TMP30]])
 ; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i16 [[TMP4]] to <16 x i1>
 ; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i16 [[X3]] to <16 x i1>
 ; CHECK-NEXT:    [[TMP14:%.*]] = select <16 x i1> [[TMP13]], <16 x i32> [[_MSPROP1]], <16 x i32> [[TMP2]]
@@ -324,10 +324,10 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssd_512(<16 x i
 ; CHECK-NEXT:    [[TMP17:%.*]] = or <16 x i32> [[TMP16]], [[TMP2]]
 ; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <16 x i1> [[TMP12]], <16 x i32> [[TMP17]], <16 x i32> [[TMP14]]
 ; CHECK-NEXT:    [[TMP18:%.*]] = select <16 x i1> [[TMP13]], <16 x i32> [[TMP11]], <16 x i32> [[X0]]
-; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
-; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <16 x i32> [[X4]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP39:%.*]] = bitcast <16 x i32> [[TMP3]] to <32 x i16>
+; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP40:%.*]] = bitcast <16 x i32> [[TMP5]] to <32 x i16>
+; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <16 x i32> [[X4]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP41:%.*]] = icmp ne <32 x i16> [[TMP39]], zeroinitializer
 ; CHECK-NEXT:    [[TMP42:%.*]] = icmp ne <32 x i16> [[TMP40]], zeroinitializer
 ; CHECK-NEXT:    [[TMP43:%.*]] = icmp ne <32 x i16> [[TMP37]], zeroinitializer
@@ -342,7 +342,7 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssd_512(<16 x i
 ; CHECK-NEXT:    [[TMP52:%.*]] = icmp ne <16 x i32> [[TMP51]], zeroinitializer
 ; CHECK-NEXT:    [[TMP53:%.*]] = sext <16 x i1> [[TMP52]] to <16 x i32>
 ; CHECK-NEXT:    [[_MSPROP3:%.*]] = or <16 x i32> [[TMP53]], [[TMP2]]
-; CHECK-NEXT:    [[TMP19:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> [[X0]], <16 x i32> [[X1]], <16 x i32> [[X4]])
+; CHECK-NEXT:    [[TMP19:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssd.512(<16 x i32> [[X0]], <32 x i16> [[TMP37]], <32 x i16> [[TMP38]])
 ; CHECK-NEXT:    [[TMP20:%.*]] = bitcast i16 [[TMP4]] to <16 x i1>
 ; CHECK-NEXT:    [[TMP21:%.*]] = bitcast i16 [[X3]] to <16 x i1>
 ; CHECK-NEXT:    [[TMP22:%.*]] = select <16 x i1> [[TMP21]], <16 x i32> [[_MSPROP3]], <16 x i32> zeroinitializer
@@ -379,10 +379,10 @@ define <16 x i32>@test_int_x86_avx512_ask_vpdpwssds_512(<16 x i32> %x0, <16 x i3
 ; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 128), align 8
 ; CHECK-NEXT:    [[TMP21:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
-; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <16 x i32> [[X2]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <16 x i32> [[TMP2]] to <32 x i16>
+; CHECK-NEXT:    [[TMP22:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <16 x i32> [[TMP3]] to <32 x i16>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <16 x i32> [[X2]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne <32 x i16> [[TMP6]], zeroinitializer
 ; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <32 x i16> [[TMP7]], zeroinitializer
 ; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne <32 x i16> [[TMP22]], zeroinitializer
@@ -397,7 +397,7 @@ define <16 x i32>@test_int_x86_avx512_ask_vpdpwssds_512(<16 x i32> %x0, <16 x i3
 ; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <16 x i32> [[TMP18]], zeroinitializer
 ; CHECK-NEXT:    [[TMP20:%.*]] = sext <16 x i1> [[TMP19]] to <16 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[TMP20]], [[TMP21]]
-; CHECK-NEXT:    [[TMP4:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> [[X0]], <16 x i32> [[X1]], <16 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP4:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> [[X0]], <32 x i16> [[TMP22]], <32 x i16> [[TMP5]])
 ; CHECK-NEXT:    store <16 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <16 x i32> [[TMP4]]
 ;
@@ -425,10 +425,10 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssds_512(<16 x
 ; CHECK-NEXT:    [[TMP9:%.*]] = xor i64 [[TMP8]], 87960930222080
 ; CHECK-NEXT:    [[TMP10:%.*]] = inttoptr i64 [[TMP9]] to ptr
 ; CHECK-NEXT:    [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP10]], align 64
-; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
-; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <16 x i32> [[X2]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP31:%.*]] = bitcast <16 x i32> [[TMP3]] to <32 x i16>
+; CHECK-NEXT:    [[TMP29:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP32:%.*]] = bitcast <16 x i32> [[_MSLD]] to <32 x i16>
+; CHECK-NEXT:    [[TMP30:%.*]] = bitcast <16 x i32> [[X2]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP33:%.*]] = icmp ne <32 x i16> [[TMP31]], zeroinitializer
 ; CHECK-NEXT:    [[TMP34:%.*]] = icmp ne <32 x i16> [[TMP32]], zeroinitializer
 ; CHECK-NEXT:    [[TMP35:%.*]] = icmp ne <32 x i16> [[TMP29]], zeroinitializer
@@ -443,7 +443,7 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssds_512(<16 x
 ; CHECK-NEXT:    [[TMP61:%.*]] = icmp ne <16 x i32> [[TMP60]], zeroinitializer
 ; CHECK-NEXT:    [[TMP62:%.*]] = sext <16 x i1> [[TMP61]] to <16 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[TMP62]], [[TMP2]]
-; CHECK-NEXT:    [[TMP11:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> [[X0]], <16 x i32> [[X1]], <16 x i32> [[X2]])
+; CHECK-NEXT:    [[TMP11:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> [[X0]], <32 x i16> [[TMP29]], <32 x i16> [[TMP30]])
 ; CHECK-NEXT:    [[TMP12:%.*]] = bitcast i16 [[TMP4]] to <16 x i1>
 ; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i16 [[X3]] to <16 x i1>
 ; CHECK-NEXT:    [[TMP14:%.*]] = select <16 x i1> [[TMP13]], <16 x i32> [[_MSPROP1]], <16 x i32> [[TMP2]]
@@ -452,10 +452,10 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssds_512(<16 x
 ; CHECK-NEXT:    [[TMP17:%.*]] = or <16 x i32> [[TMP16]], [[TMP2]]
 ; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <16 x i1> [[TMP12]], <16 x i32> [[TMP17]], <16 x i32> [[TMP14]]
 ; CHECK-NEXT:    [[TMP18:%.*]] = select <16 x i1> [[TMP13]], <16 x i32> [[TMP11]], <16 x i32> [[X0]]
-; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
-; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <16 x i32> [[X4]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP39:%.*]] = bitcast <16 x i32> [[TMP3]] to <32 x i16>
+; CHECK-NEXT:    [[TMP37:%.*]] = bitcast <16 x i32> [[X1]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP40:%.*]] = bitcast <16 x i32> [[TMP5]] to <32 x i16>
+; CHECK-NEXT:    [[TMP38:%.*]] = bitcast <16 x i32> [[X4]] to <32 x i16>
 ; CHECK-NEXT:    [[TMP41:%.*]] = icmp ne <32 x i16> [[TMP39]], zeroinitializer
 ; CHECK-NEXT:    [[TMP42:%.*]] = icmp ne <32 x i16> [[TMP40]], zeroinitializer
 ; CHECK-NEXT:    [[TMP43:%.*]] = icmp ne <32 x i16> [[TMP37]], zeroinitializer
@@ -470,7 +470,7 @@ define { <16 x i32>, <16 x i32> } @test_int_x86_avx512_mask_vpdpwssds_512(<16 x
 ; CHECK-NEXT:    [[TMP52:%.*]] = icmp ne <16 x i32> [[TMP51]], zeroinitializer
 ; CHECK-NEXT:    [[TMP53:%.*]] = sext <16 x i1> [[TMP52]] to <16 x i32>
 ; CHECK-NEXT:    [[_MSPROP3:%.*]] = or <16 x i32> [[TMP53]], [[TMP2]]
-; CHECK-NEXT:    [[TMP19:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> [[X0]], <16 x i32> [[X1]], <16 x i32> [[X4]])
+; CHECK-NEXT:    [[TMP19:%.*]] = call <16 x i32> @llvm.x86.avx512.vpdpwssds.512(<16 x i32> [[X0]], <32 x i16> [[TMP37]], <32 x i16> [[TMP38]])
 ; CHECK-NEXT:    [[TMP20:%.*]] = bitcast i16 [[TMP4]] to <16 x i1>
 ; CHECK-NEXT:    [[TMP21:%.*]] = bitcast i16 [[X3]] to <16 x i1>
 ; CHECK-NEXT:    [[TMP22:%.*]] = select <16 x i1> [[TMP21]], <16 x i32> [[_MSPROP3]], <16 x i32> zeroinitializer
diff --git a/llvm/test/Instrumentation/MemorySanitizer/X86/avx_vnni-intrinsics.ll b/llvm/test/Instrumentation/MemorySanitizer/X86/avx_vnni-intrinsics.ll
index 0344fbd5ee2a9..0a6c5f7b089ba 100644
--- a/llvm/test/Instrumentation/MemorySanitizer/X86/avx_vnni-intrinsics.ll
+++ b/llvm/test/Instrumentation/MemorySanitizer/X86/avx_vnni-intrinsics.ll
@@ -143,10 +143,10 @@ define <8 x i32>@test_int_x86_avx_vpdpwssd_256(<8 x i32> %x0, <8 x i32> %x1, <8
 ; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
 ; CHECK-NEXT:    [[TMP21:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
-; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <8 x i32> [[TMP2]] to <16 x i16>
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <8 x i32> [[TMP3]] to <16 x i16>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne <16 x i16> [[TMP6]], zeroinitializer
 ; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <16 x i16> [[TMP7]], zeroinitializer
 ; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne <16 x i16> [[TMP4]], zeroinitializer
@@ -161,7 +161,7 @@ define <8 x i32>@test_int_x86_avx_vpdpwssd_256(<8 x i32> %x0, <8 x i32> %x1, <8
 ; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <8 x i32> [[TMP18]], zeroinitializer
 ; CHECK-NEXT:    [[TMP20:%.*]] = sext <8 x i1> [[TMP19]] to <8 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[TMP20]], [[TMP21]]
-; CHECK-NEXT:    [[RES:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X2]])
+; CHECK-NEXT:    [[RES:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> [[X0]], <16 x i16> [[TMP4]], <16 x i16> [[TMP5]])
 ; CHECK-NEXT:    store <8 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <8 x i32> [[RES]]
 ;
@@ -178,10 +178,10 @@ define <4 x i32>@test_int_x86_avx_vpdpwssd_128(<4 x i32> %x0, <4 x i32> %x1, <4
 ; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
 ; CHECK-NEXT:    [[TMP21:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
-; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <4 x i32> [[TMP2]] to <8 x i16>
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <4 x i32> [[TMP3]] to <8 x i16>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne <8 x i16> [[TMP6]], zeroinitializer
 ; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <8 x i16> [[TMP7]], zeroinitializer
 ; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne <8 x i16> [[TMP4]], zeroinitializer
@@ -196,7 +196,7 @@ define <4 x i32>@test_int_x86_avx_vpdpwssd_128(<4 x i32> %x0, <4 x i32> %x1, <4
 ; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <4 x i32> [[TMP18]], zeroinitializer
 ; CHECK-NEXT:    [[TMP20:%.*]] = sext <4 x i1> [[TMP19]] to <4 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[TMP20]], [[TMP21]]
-; CHECK-NEXT:    [[RES:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X2]])
+; CHECK-NEXT:    [[RES:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> [[X0]], <8 x i16> [[TMP4]], <8 x i16> [[TMP5]])
 ; CHECK-NEXT:    store <4 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <4 x i32> [[RES]]
 ;
@@ -213,10 +213,10 @@ define <8 x i32>@test_int_x86_avx_vpdpwssds_256(<8 x i32> %x0, <8 x i32> %x1, <8
 ; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
 ; CHECK-NEXT:    [[TMP21:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
-; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <8 x i32> [[TMP2]] to <16 x i16>
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <8 x i32> [[X1]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <8 x i32> [[TMP3]] to <16 x i16>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <8 x i32> [[X2]] to <16 x i16>
 ; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne <16 x i16> [[TMP6]], zeroinitializer
 ; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <16 x i16> [[TMP7]], zeroinitializer
 ; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne <16 x i16> [[TMP4]], zeroinitializer
@@ -231,7 +231,7 @@ define <8 x i32>@test_int_x86_avx_vpdpwssds_256(<8 x i32> %x0, <8 x i32> %x1, <8
 ; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <8 x i32> [[TMP18]], zeroinitializer
 ; CHECK-NEXT:    [[TMP20:%.*]] = sext <8 x i1> [[TMP19]] to <8 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[TMP20]], [[TMP21]]
-; CHECK-NEXT:    [[RES:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X2]])
+; CHECK-NEXT:    [[RES:%.*]] = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> [[X0]], <16 x i16> [[TMP4]], <16 x i16> [[TMP5]])
 ; CHECK-NEXT:    store <8 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <8 x i32> [[RES]]
 ;
@@ -248,10 +248,10 @@ define <4 x i32>@test_int_x86_avx_vpdpwssds_128(<4 x i32> %x0, <4 x i32> %x1, <4
 ; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
 ; CHECK-NEXT:    [[TMP21:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
-; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP6:%.*]] = bitcast <4 x i32> [[TMP2]] to <8 x i16>
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <4 x i32> [[X1]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <4 x i32> [[TMP3]] to <8 x i16>
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <4 x i32> [[X2]] to <8 x i16>
 ; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne <8 x i16> [[TMP6]], zeroinitializer
 ; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <8 x i16> [[TMP7]], zeroinitializer
 ; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne <8 x i16> [[TMP4]], zeroinitializer
@@ -266,7 +266,7 @@ define <4 x i32>@test_int_x86_avx_vpdpwssds_128(<4 x i32> %x0, <4 x i32> %x1, <4
 ; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <4 x i32> [[TMP18]], zeroinitializer
 ; CHECK-NEXT:    [[TMP20:%.*]] = sext <4 x i1> [[TMP19]] to <4 x i32>
 ; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[TMP20]], [[TMP21]]
-; CHECK-NEXT:    [[RES:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X2]])
+; CHECK-NEXT:    [[RES:%.*]] = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> [[X0]], <8 x i16> [[TMP4]], <8 x i16> [[TMP5]])
 ; CHECK-NEXT:    store <4 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <4 x i32> [[RES]]
 ;

>From cc40d7b6af5e8fd09433d881a2ef841c95c25333 Mon Sep 17 00:00:00 2001
From: BaiXilin <x53bai at uwaterloo.ca>
Date: Sun, 30 Nov 2025 23:58:09 -0500
Subject: [PATCH 2/4]  - vpdpwsud_128, vpdpwsud_256, vpdowsud_512  -
 vpdpwsuds_128, vpdpwsuds_256, vpdpwsuds_512  - vpdpwusd_128, vpdpwusd_256,
 vpdpwusd_512  - vpdpwusds_128, vpdpwusds_256, vpdpwusds_512  - vpdpwuud_128,
 vpdpwuud_256, vpdpwuud_512  - vpdpwuuds_128, vpdpwuuds_256, vpdpwuuds_512

---
 clang/include/clang/Basic/BuiltinsX86.td      |  36 +-
 clang/lib/Headers/avx10_2_512niintrin.h       |  24 +-
 clang/lib/Headers/avxvnniint16intrin.h        |  97 +++--
 .../test/CodeGen/X86/avx10_2_512ni-builtins.c |  36 +-
 clang/test/CodeGen/X86/avx10_2ni-builtins.c   |  48 +--
 .../test/CodeGen/X86/avxvnniint16-builtins.c  |  24 +-
 llvm/include/llvm/IR/IntrinsicsX86.td         |  36 +-
 llvm/lib/IR/AutoUpgrade.cpp                   | 111 ++++--
 .../Instrumentation/MemorySanitizer.cpp       |  66 +++
 .../CodeGen/X86/avx10.2-intrinsic-upgrade.ll  |  96 +++++
 .../CodeGen/X86/avx10_2_512ni-intrinsics.ll   |  18 +-
 llvm/test/CodeGen/X86/avx10_2ni-intrinsics.ll |  24 +-
 .../X86/avxvnniint16-intrinsics-upgrade.ll    | 185 +++++++++
 .../CodeGen/X86/avxvnniint16-intrinsics.ll    |  72 ++--
 .../X86/stack-folding-int-avxvnniint16.ll     |  24 +-
 .../X86/avx10_2_512ni-intrinsics.ll           | 290 +++++++++-----
 .../X86/avx10_2ni-intrinsics.ll               | 360 ++++++++++++-----
 .../X86/avxvnniint16-intrinsics.ll            | 376 ++++++++++++------
 18 files changed, 1371 insertions(+), 552 deletions(-)
 create mode 100644 llvm/test/CodeGen/X86/avxvnniint16-intrinsics-upgrade.ll

diff --git a/clang/include/clang/Basic/BuiltinsX86.td b/clang/include/clang/Basic/BuiltinsX86.td
index f61dd0546cc24..da34d4123628f 100644
--- a/clang/include/clang/Basic/BuiltinsX86.td
+++ b/clang/include/clang/Basic/BuiltinsX86.td
@@ -4325,12 +4325,12 @@ let Features = "avx10.2", Attributes = [NoThrow, Const, RequiredVectorWidth<512>
 }
 
 let Features = "avx10.2", Attributes = [NoThrow, RequiredVectorWidth<512>] in {
-  def vpdpwsud512 : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<16, int>, _Vector<16, int>)">;
-  def vpdpwsuds512 : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<16, int>, _Vector<16, int>)">;
-  def vpdpwusd512 : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<16, int>, _Vector<16, int>)">;
-  def vpdpwusds512 : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<16, int>, _Vector<16, int>)">;
-  def vpdpwuud512 : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<16, int>, _Vector<16, int>)">;
-  def vpdpwuuds512 : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<16, int>, _Vector<16, int>)">;
+  def vpdpwsud512 : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<32, short>, _Vector<32, unsigned short>)">;
+  def vpdpwsuds512 : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<32, short>, _Vector<32, unsigned short>)">;
+  def vpdpwusd512 : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<32, unsigned short>, _Vector<32, short>)">;
+  def vpdpwusds512 : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<32, unsigned short>, _Vector<32, short>)">;
+  def vpdpwuud512 : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<32, unsigned short>, _Vector<32, unsigned short>)">;
+  def vpdpwuuds512 : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<32, unsigned short>, _Vector<32, unsigned short>)">;
 }
 
 let Features = "avx10.2", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in {
@@ -4338,51 +4338,51 @@ let Features = "avx10.2", Attributes = [NoThrow, Const, RequiredVectorWidth<512>
 }
 
 let Features = "avxvnniint16|avx10.2", Attributes = [NoThrow, RequiredVectorWidth<128>] in {
-  def vpdpwsud128 : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<4, int>, _Vector<4, int>)">;
+  def vpdpwsud128 : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<8, short>, _Vector<8, unsigned short>)">;
 }
 
 let Features = "avxvnniint16|avx10.2", Attributes = [NoThrow, RequiredVectorWidth<256>] in {
-  def vpdpwsud256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<8, int>, _Vector<8, int>)">;
+  def vpdpwsud256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<16, short>, _Vector<16, unsigned short>)">;
 }
 
 let Features = "avxvnniint16|avx10.2", Attributes = [NoThrow, RequiredVectorWidth<128>] in {
-  def vpdpwsuds128 : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<4, int>, _Vector<4, int>)">;
+  def vpdpwsuds128 : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<8, short>, _Vector<8, unsigned short>)">;
 }
 
 let Features = "avxvnniint16|avx10.2", Attributes = [NoThrow, RequiredVectorWidth<256>] in {
-  def vpdpwsuds256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<8, int>, _Vector<8, int>)">;
+  def vpdpwsuds256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<16, short>, _Vector<16, unsigned short>)">;
 }
 
 let Features = "avxvnniint16|avx10.2", Attributes = [NoThrow, RequiredVectorWidth<128>] in {
-  def vpdpwusd128 : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<4, int>, _Vector<4, int>)">;
+  def vpdpwusd128 : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<8, unsigned short>, _Vector<8, short>)">;
 }
 
 let Features = "avxvnniint16|avx10.2", Attributes = [NoThrow, RequiredVectorWidth<256>] in {
-  def vpdpwusd256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<8, int>, _Vector<8, int>)">;
+  def vpdpwusd256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<16, unsigned short>, _Vector<16, short>)">;
 }
 
 let Features = "avxvnniint16|avx10.2", Attributes = [NoThrow, RequiredVectorWidth<128>] in {
-  def vpdpwusds128 : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<4, int>, _Vector<4, int>)">;
+  def vpdpwusds128 : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<8, unsigned short>, _Vector<8, short>)">;
 }
 
 let Features = "avxvnniint16|avx10.2", Attributes = [NoThrow, RequiredVectorWidth<256>] in {
-  def vpdpwusds256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<8, int>, _Vector<8, int>)">;
+  def vpdpwusds256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<16, unsigned short>, _Vector<16, short>)">;
 }
 
 let Features = "avxvnniint16|avx10.2", Attributes = [NoThrow, RequiredVectorWidth<128>] in {
-  def vpdpwuud128 : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<4, int>, _Vector<4, int>)">;
+  def vpdpwuud128 : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<8, unsigned short>, _Vector<8, unsigned short>)">;
 }
 
 let Features = "avxvnniint16|avx10.2", Attributes = [NoThrow, RequiredVectorWidth<256>] in {
-  def vpdpwuud256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<8, int>, _Vector<8, int>)">;
+  def vpdpwuud256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<16, unsigned short>, _Vector<16, unsigned short>)">;
 }
 
 let Features = "avxvnniint16|avx10.2", Attributes = [NoThrow, RequiredVectorWidth<128>] in {
-  def vpdpwuuds128 : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<4, int>, _Vector<4, int>)">;
+  def vpdpwuuds128 : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<8, unsigned short>, _Vector<8, unsigned short>)">;
 }
 
 let Features = "avxvnniint16|avx10.2", Attributes = [NoThrow, RequiredVectorWidth<256>] in {
-  def vpdpwuuds256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<8, int>, _Vector<8, int>)">;
+  def vpdpwuuds256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<16, unsigned short>, _Vector<16, unsigned short>)">;
 }
 
 let Features = "avx10.2", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in {
diff --git a/clang/lib/Headers/avx10_2_512niintrin.h b/clang/lib/Headers/avx10_2_512niintrin.h
index fdb57c7c9e27b..b2215b72c57bc 100644
--- a/clang/lib/Headers/avx10_2_512niintrin.h
+++ b/clang/lib/Headers/avx10_2_512niintrin.h
@@ -185,8 +185,8 @@ static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_maskz_dpbuuds_epi32(
 static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_dpwsud_epi32(__m512i __A,
                                                                  __m512i __B,
                                                                  __m512i __C) {
-  return (__m512i)__builtin_ia32_vpdpwsud512((__v16si)__A, (__v16si)__B,
-                                             (__v16si)__C);
+  return (__m512i)__builtin_ia32_vpdpwsud512((__v16si)__A, (__v32hi)__B,
+                                             (__v32hu)__C);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
@@ -206,8 +206,8 @@ static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_maskz_dpwsud_epi32(
 static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_dpwsuds_epi32(__m512i __A,
                                                                   __m512i __B,
                                                                   __m512i __C) {
-  return (__m512i)__builtin_ia32_vpdpwsuds512((__v16si)__A, (__v16si)__B,
-                                              (__v16si)__C);
+  return (__m512i)__builtin_ia32_vpdpwsuds512((__v16si)__A, (__v32hi)__B,
+                                              (__v32hu)__C);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_mask_dpwsuds_epi32(
@@ -227,8 +227,8 @@ static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_maskz_dpwsuds_epi32(
 static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_dpwusd_epi32(__m512i __A,
                                                                  __m512i __B,
                                                                  __m512i __C) {
-  return (__m512i)__builtin_ia32_vpdpwusd512((__v16si)__A, (__v16si)__B,
-                                             (__v16si)__C);
+  return (__m512i)__builtin_ia32_vpdpwusd512((__v16si)__A, (__v32hu)__B,
+                                             (__v32hi)__C);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
@@ -248,8 +248,8 @@ static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_maskz_dpwusd_epi32(
 static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_dpwusds_epi32(__m512i __A,
                                                                   __m512i __B,
                                                                   __m512i __C) {
-  return (__m512i)__builtin_ia32_vpdpwusds512((__v16si)__A, (__v16si)__B,
-                                              (__v16si)__C);
+  return (__m512i)__builtin_ia32_vpdpwusds512((__v16si)__A, (__v32hu)__B,
+                                              (__v32hi)__C);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_mask_dpwusds_epi32(
@@ -269,8 +269,8 @@ static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_maskz_dpwusds_epi32(
 static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_dpwuud_epi32(__m512i __A,
                                                                  __m512i __B,
                                                                  __m512i __C) {
-  return (__m512i)__builtin_ia32_vpdpwuud512((__v16si)__A, (__v16si)__B,
-                                             (__v16si)__C);
+  return (__m512i)__builtin_ia32_vpdpwuud512((__v16si)__A, (__v32hu)__B,
+                                             (__v32hu)__C);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS
@@ -290,8 +290,8 @@ static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_maskz_dpwuud_epi32(
 static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_dpwuuds_epi32(__m512i __A,
                                                                   __m512i __B,
                                                                   __m512i __C) {
-  return (__m512i)__builtin_ia32_vpdpwuuds512((__v16si)__A, (__v16si)__B,
-                                              (__v16si)__C);
+  return (__m512i)__builtin_ia32_vpdpwuuds512((__v16si)__A, (__v32hu)__B,
+                                              (__v32hu)__C);
 }
 
 static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_mask_dpwuuds_epi32(
diff --git a/clang/lib/Headers/avxvnniint16intrin.h b/clang/lib/Headers/avxvnniint16intrin.h
index 805d249911c17..801795ac543dd 100644
--- a/clang/lib/Headers/avxvnniint16intrin.h
+++ b/clang/lib/Headers/avxvnniint16intrin.h
@@ -15,6 +15,7 @@
 #ifndef __AVXVNNIINT16INTRIN_H
 #define __AVXVNNIINT16INTRIN_H
 
+// clang-format off
 /// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a __A with
 ///    corresponding unsigned 16-bit integers in \a __B, producing 2 intermediate
 ///    signed 16-bit results. Sum these 2 results with the corresponding
@@ -45,10 +46,12 @@
 /// ENDFOR
 /// dst[MAX:128] := 0
 /// \endcode
+// clang-format on
 #define _mm_dpwsud_epi32(__W, __A, __B)                                        \
-  ((__m128i)__builtin_ia32_vpdpwsud128((__v4si)(__W), (__v4si)(__A),           \
-                                       (__v4si)(__B)))
+  ((__m128i)__builtin_ia32_vpdpwsud128((__v4si)(__W), (__v8hi)(__A),           \
+                                       (__v8hu)(__B)))
 
+// clang-format off
 /// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a __A with
 ///    corresponding unsigned 16-bit integers in \a __B, producing 2 intermediate
 ///    signed 16-bit results. Sum these 2 results with the corresponding
@@ -79,10 +82,12 @@
 /// ENDFOR
 /// dst[MAX:256] := 0
 /// \endcode
+// clang-format on
 #define _mm256_dpwsud_epi32(__W, __A, __B)                                     \
-  ((__m256i)__builtin_ia32_vpdpwsud256((__v8si)(__W), (__v8si)(__A),           \
-                                       (__v8si)(__B)))
+  ((__m256i)__builtin_ia32_vpdpwsud256((__v8si)(__W), (__v16hi)(__A),          \
+                                       (__v16hu)(__B)))
 
+// clang-format off
 /// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a __A with
 ///    corresponding unsigned 16-bit integers in \a __B, producing 2 intermediate
 ///    signed 16-bit results. Sum these 2 results with the corresponding
@@ -114,10 +119,13 @@
 /// ENDFOR
 /// dst[MAX:128] := 0
 /// \endcode
+// clang-format on
+/// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a __A with
 #define _mm_dpwsuds_epi32(__W, __A, __B)                                       \
-  ((__m128i)__builtin_ia32_vpdpwsuds128((__v4si)(__W), (__v4si)(__A),          \
-                                        (__v4si)(__B)))
+  ((__m128i)__builtin_ia32_vpdpwsuds128((__v4si)(__W), (__v8hi)(__A),          \
+                                        (__v8hu)(__B)))
 
+// clang-format off
 /// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a __A with
 ///    corresponding unsigned 16-bit integers in \a __B, producing 2 intermediate
 ///    signed 16-bit results. Sum these 2 results with the corresponding
@@ -149,10 +157,12 @@
 /// ENDFOR
 /// dst[MAX:256] := 0
 /// \endcode
+// clang-format on
 #define _mm256_dpwsuds_epi32(__W, __A, __B)                                    \
-  ((__m256i)__builtin_ia32_vpdpwsuds256((__v8si)(__W), (__v8si)(__A),          \
-                                        (__v8si)(__B)))
+  ((__m256i)__builtin_ia32_vpdpwsuds256((__v8si)(__W), (__v16hi)(__A),         \
+                                        (__v16hu)(__B)))
 
+// clang-format off
 /// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A with
 ///    corresponding signed 16-bit integers in \a __B, producing 2 intermediate
 ///    signed 16-bit results. Sum these 2 results with the corresponding
@@ -183,10 +193,12 @@
 /// ENDFOR
 /// dst[MAX:128] := 0
 /// \endcode
+// clang-format on
 #define _mm_dpwusd_epi32(__W, __A, __B)                                        \
-  ((__m128i)__builtin_ia32_vpdpwusd128((__v4si)(__W), (__v4si)(__A),           \
-                                       (__v4si)(__B)))
+  ((__m128i)__builtin_ia32_vpdpwusd128((__v4si)(__W), (__v8hu)(__A),           \
+                                       (__v8hi)(__B)))
 
+// clang-format off
 /// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A with
 ///    corresponding signed 16-bit integers in \a __B, producing 2 intermediate
 ///    signed 16-bit results. Sum these 2 results with the corresponding
@@ -217,10 +229,12 @@
 /// ENDFOR
 /// dst[MAX:256] := 0
 /// \endcode
+// clang-format on
 #define _mm256_dpwusd_epi32(__W, __A, __B)                                     \
-  ((__m256i)__builtin_ia32_vpdpwusd256((__v8si)(__W), (__v8si)(__A),           \
-                                       (__v8si)(__B)))
+  ((__m256i)__builtin_ia32_vpdpwusd256((__v8si)(__W), (__v16hu)(__A),          \
+                                       (__v16hi)(__B)))
 
+// clang-format off
 /// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A with
 ///    corresponding signed 16-bit integers in \a __B, producing 2 intermediate
 ///    signed 16-bit results. Sum these 2 results with the corresponding
@@ -233,7 +247,7 @@
 /// __m128i _mm_dpwusds_epi32(__m128i __W, __m128i __A, __m128i __B)
 /// \endcode
 ///
-/// This intrinsic corresponds to the \c VPDPWSUDS instruction.
+/// This intrinsic corresponds to the \c VPDPWUSDS instruction.
 ///
 /// \param __W
 ///    A 128-bit vector of [4 x int].
@@ -252,10 +266,12 @@
 /// ENDFOR
 /// dst[MAX:128] := 0
 /// \endcode
+// clang-format on
 #define _mm_dpwusds_epi32(__W, __A, __B)                                       \
-  ((__m128i)__builtin_ia32_vpdpwusds128((__v4si)(__W), (__v4si)(__A),          \
-                                        (__v4si)(__B)))
+  ((__m128i)__builtin_ia32_vpdpwusds128((__v4si)(__W), (__v8hu)(__A),          \
+                                        (__v8hi)(__B)))
 
+// clang-format off
 /// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A with
 ///    corresponding signed 16-bit integers in \a __B, producing 2 intermediate
 ///    signed 16-bit results. Sum these 2 results with the corresponding
@@ -268,7 +284,7 @@
 /// __m256i _mm256_dpwsuds_epi32(__m256i __W, __m256i __A, __m256i __B)
 /// \endcode
 ///
-/// This intrinsic corresponds to the \c VPDPWSUDS instruction.
+/// This intrinsic corresponds to the \c VPDPWUSDS instruction.
 ///
 /// \param __W
 ///    A 256-bit vector of [8 x int].
@@ -287,10 +303,12 @@
 /// ENDFOR
 /// dst[MAX:256] := 0
 /// \endcode
+// clang-format on
 #define _mm256_dpwusds_epi32(__W, __A, __B)                                    \
-  ((__m256i)__builtin_ia32_vpdpwusds256((__v8si)(__W), (__v8si)(__A),          \
-                                        (__v8si)(__B)))
+  ((__m256i)__builtin_ia32_vpdpwusds256((__v8si)(__W), (__v16hu)(__A),         \
+                                        (__v16hi)(__B)))
 
+// clang-format off
 /// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A with
 ///    corresponding unsigned 16-bit integers in \a __B, producing 2 intermediate
 ///    signed 16-bit results. Sum these 2 results with the corresponding
@@ -305,13 +323,13 @@
 /// This intrinsic corresponds to the \c VPDPWUUD instruction.
 ///
 /// \param __W
-///    A 128-bit vector of [4 x unsigned int].
+///    A 128-bit vector of [4 x int].
 /// \param __A
 ///    A 128-bit vector of [8 x unsigned short].
 /// \param __B
 ///    A 128-bit vector of [8 x unsigned short].
 /// \returns
-///    A 128-bit vector of [4 x unsigned int].
+///    A 128-bit vector of [4 x int].
 ///
 /// \code{.operation}
 /// FOR j := 0 to 3
@@ -321,10 +339,12 @@
 /// ENDFOR
 /// dst[MAX:128] := 0
 /// \endcode
+// clang-format on
 #define _mm_dpwuud_epi32(__W, __A, __B)                                        \
-  ((__m128i)__builtin_ia32_vpdpwuud128((__v4si)(__W), (__v4si)(__A),           \
-                                       (__v4si)(__B)))
+  ((__m128i)__builtin_ia32_vpdpwuud128((__v4si)(__W), (__v8hu)(__A),           \
+                                       (__v8hu)(__B)))
 
+// clang-format off
 /// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A with
 ///    corresponding unsigned 16-bit integers in \a __B, producing 2 intermediate
 ///    signed 16-bit results. Sum these 2 results with the corresponding
@@ -339,13 +359,13 @@
 /// This intrinsic corresponds to the \c VPDPWUUD instruction.
 ///
 /// \param __W
-///    A 256-bit vector of [8 x unsigned int].
+///    A 256-bit vector of [8 x int].
 /// \param __A
 ///    A 256-bit vector of [16 x unsigned short].
 /// \param __B
 ///    A 256-bit vector of [16 x unsigned short].
 /// \returns
-///    A 256-bit vector of [8 x unsigned int].
+///    A 256-bit vector of [8 x int].
 ///
 /// \code{.operation}
 /// FOR j := 0 to 7
@@ -355,10 +375,12 @@
 /// ENDFOR
 /// dst[MAX:256] := 0
 /// \endcode
+// clang-format on
 #define _mm256_dpwuud_epi32(__W, __A, __B)                                     \
-  ((__m256i)__builtin_ia32_vpdpwuud256((__v8si)(__W), (__v8si)(__A),           \
-                                       (__v8si)(__B)))
+  ((__m256i)__builtin_ia32_vpdpwuud256((__v8si)(__W), (__v16hu)(__A),          \
+                                       (__v16hu)(__B)))
 
+// clang-format off
 /// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A with
 ///    corresponding unsigned 16-bit integers in \a __B, producing 2 intermediate
 ///    signed 16-bit results. Sum these 2 results with the corresponding
@@ -371,16 +393,16 @@
 /// __m128i _mm_dpwsuds_epi32(__m128i __W, __m128i __A, __m128i __B)
 /// \endcode
 ///
-/// This intrinsic corresponds to the \c VPDPWSUDS instruction.
+/// This intrinsic corresponds to the \c VPDPWUUDS instruction.
 ///
 /// \param __W
-///    A 128-bit vector of [4 x unsigned int].
+///    A 128-bit vector of [4 x int].
 /// \param __A
 ///    A 128-bit vector of [8 x unsigned short].
 /// \param __B
 ///    A 128-bit vector of [8 x unsigned short].
 /// \returns
-///    A 128-bit vector of [4 x unsigned int].
+///    A 128-bit vector of [4 x int].
 ///
 /// \code{.operation}
 /// FOR j := 0 to 3
@@ -390,10 +412,12 @@
 /// ENDFOR
 /// dst[MAX:128] := 0
 /// \endcode
+// clang-format on
 #define _mm_dpwuuds_epi32(__W, __A, __B)                                       \
-  ((__m128i)__builtin_ia32_vpdpwuuds128((__v4si)(__W), (__v4si)(__A),          \
-                                        (__v4si)(__B)))
+  ((__m128i)__builtin_ia32_vpdpwuuds128((__v4si)(__W), (__v8hu)(__A),          \
+                                        (__v8hu)(__B)))
 
+// clang-format off
 /// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A with
 ///    corresponding unsigned 16-bit integers in \a __B, producing 2 intermediate
 ///    signed 16-bit results. Sum these 2 results with the corresponding
@@ -406,16 +430,16 @@
 /// __m256i _mm256_dpwuuds_epi32(__m256i __W, __m256i __A, __m256i __B)
 /// \endcode
 ///
-/// This intrinsic corresponds to the \c VPDPWSUDS instruction.
+/// This intrinsic corresponds to the \c VPDPWUUDS instruction.
 ///
 /// \param __W
-///    A 256-bit vector of [8 x unsigned int].
+///    A 256-bit vector of [8 x int].
 /// \param __A
 ///    A 256-bit vector of [16 x unsigned short].
 /// \param __B
 ///    A 256-bit vector of [16 x unsigned short].
 /// \returns
-///    A 256-bit vector of [8 x unsigned int].
+///    A 256-bit vector of [8 x int].
 ///
 /// \code{.operation}
 /// FOR j := 0 to 7
@@ -425,8 +449,9 @@
 /// ENDFOR
 /// dst[MAX:256] := 0
 /// \endcode
+// clang-format on
 #define _mm256_dpwuuds_epi32(__W, __A, __B)                                    \
-  ((__m256i)__builtin_ia32_vpdpwuuds256((__v8si)(__W), (__v8si)(__A),          \
-                                        (__v8si)(__B)))
+  ((__m256i)__builtin_ia32_vpdpwuuds256((__v8si)(__W), (__v16hu)(__A),         \
+                                        (__v16hu)(__B)))
 
 #endif // __AVXVNNIINT16INTRIN_H
diff --git a/clang/test/CodeGen/X86/avx10_2_512ni-builtins.c b/clang/test/CodeGen/X86/avx10_2_512ni-builtins.c
index 728c9f5652dd9..1ba6e87653a74 100644
--- a/clang/test/CodeGen/X86/avx10_2_512ni-builtins.c
+++ b/clang/test/CodeGen/X86/avx10_2_512ni-builtins.c
@@ -176,20 +176,20 @@ __m512i test_mm512_maskz_dpbuuds_epi32(__mmask16 __U, __m512i __W, __m512i __A,
 /* VNNI INT16 */
 __m512i test_mm512_dpwsud_epi32(__m512i __A, __m512i __B, __m512i __C) {
 // CHECK-LABEL: @test_mm512_dpwsud_epi32(
-// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
   return _mm512_dpwsud_epi32(__A, __B, __C);
 }
 
 __m512i test_mm512_mask_dpwsud_epi32(__m512i __A, __mmask16 __B, __m512i __C, __m512i __D) {
 // CHECK-LABEL: @test_mm512_mask_dpwsud_epi32(
-// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
 // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
   return _mm512_mask_dpwsud_epi32(__A, __B, __C, __D);
 }
 
 __m512i test_mm512_maskz_dpwsud_epi32(__mmask16 __U, __m512i __A, __m512i __B, __m512i __C) {
 // CHECK-LABEL: @test_mm512_maskz_dpwsud_epi32(
-// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
 // CHECK: zeroinitializer
 // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
   return _mm512_maskz_dpwsud_epi32(__U, __A, __B, __C);
@@ -197,20 +197,20 @@ __m512i test_mm512_maskz_dpwsud_epi32(__mmask16 __U, __m512i __A, __m512i __B, _
 
 __m512i test_mm512_dpwsuds_epi32(__m512i __A, __m512i __B, __m512i __C) {
 // CHECK-LABEL: @test_mm512_dpwsuds_epi32(
-// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwsuds.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwsuds.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
   return _mm512_dpwsuds_epi32(__A, __B, __C);
 }
 
 __m512i test_mm512_mask_dpwsuds_epi32(__m512i __A, __mmask16 __B, __m512i __C, __m512i __D) {
 // CHECK-LABEL: @test_mm512_mask_dpwsuds_epi32(
-// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwsuds.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwsuds.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
 // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
   return _mm512_mask_dpwsuds_epi32(__A, __B, __C, __D);
 }
 
 __m512i test_mm512_maskz_dpwsuds_epi32(__mmask16 __U, __m512i __A, __m512i __B, __m512i __C) {
 // CHECK-LABEL: @test_mm512_maskz_dpwsuds_epi32(
-// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwsuds.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwsuds.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
 // CHECK: zeroinitializer
 // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
   return _mm512_maskz_dpwsuds_epi32(__U, __A, __B, __C);
@@ -218,20 +218,20 @@ __m512i test_mm512_maskz_dpwsuds_epi32(__mmask16 __U, __m512i __A, __m512i __B,
 
 __m512i test_mm512_dpwusd_epi32(__m512i __A, __m512i __B, __m512i __C) {
 // CHECK-LABEL: @test_mm512_dpwusd_epi32(
-// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwusd.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwusd.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
   return _mm512_dpwusd_epi32(__A, __B, __C);
 }
 
 __m512i test_mm512_mask_dpwusd_epi32(__m512i __A, __mmask16 __B, __m512i __C, __m512i __D) {
 // CHECK-LABEL: @test_mm512_mask_dpwusd_epi32(
-// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwusd.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwusd.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
 // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
   return _mm512_mask_dpwusd_epi32(__A, __B, __C, __D);
 }
 
 __m512i test_mm512_maskz_dpwusd_epi32(__mmask16 __U, __m512i __A, __m512i __B, __m512i __C) {
 // CHECK-LABEL: @test_mm512_maskz_dpwusd_epi32(
-// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwusd.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwusd.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
 // CHECK: zeroinitializer
 // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
   return _mm512_maskz_dpwusd_epi32(__U, __A, __B, __C);
@@ -239,20 +239,20 @@ __m512i test_mm512_maskz_dpwusd_epi32(__mmask16 __U, __m512i __A, __m512i __B, _
 
 __m512i test_mm512_dpwusds_epi32(__m512i __A, __m512i __B, __m512i __C) {
 // CHECK-LABEL: @test_mm512_dpwusds_epi32(
-// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwusds.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwusds.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
   return _mm512_dpwusds_epi32(__A, __B, __C);
 }
 
 __m512i test_mm512_mask_dpwusds_epi32(__m512i __A, __mmask16 __B, __m512i __C, __m512i __D) {
 // CHECK-LABEL: @test_mm512_mask_dpwusds_epi32(
-// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwusds.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwusds.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
 // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
   return _mm512_mask_dpwusds_epi32(__A, __B, __C, __D);
 }
 
 __m512i test_mm512_maskz_dpwusds_epi32(__mmask16 __U, __m512i __A, __m512i __B, __m512i __C) {
 // CHECK-LABEL: @test_mm512_maskz_dpwusds_epi32(
-// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwusds.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwusds.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
 // CHECK: zeroinitializer
 // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
   return _mm512_maskz_dpwusds_epi32(__U, __A, __B, __C);
@@ -260,20 +260,20 @@ __m512i test_mm512_maskz_dpwusds_epi32(__mmask16 __U, __m512i __A, __m512i __B,
 
 __m512i test_mm512_dpwuud_epi32(__m512i __A, __m512i __B, __m512i __C) {
 // CHECK-LABEL: @test_mm512_dpwuud_epi32(
-// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwuud.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwuud.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
   return _mm512_dpwuud_epi32(__A, __B, __C);
 }
 
 __m512i test_mm512_mask_dpwuud_epi32(__m512i __A, __mmask16 __B, __m512i __C, __m512i __D) {
 // CHECK-LABEL: @test_mm512_mask_dpwuud_epi32(
-// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwuud.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwuud.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
 // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
   return _mm512_mask_dpwuud_epi32(__A, __B, __C, __D);
 }
 
 __m512i test_mm512_maskz_dpwuud_epi32(__mmask16 __U, __m512i __A, __m512i __B, __m512i __C) {
 // CHECK-LABEL: @test_mm512_maskz_dpwuud_epi32(
-// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwuud.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwuud.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
 // CHECK: zeroinitializer
 // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
   return _mm512_maskz_dpwuud_epi32(__U, __A, __B, __C);
@@ -281,20 +281,20 @@ __m512i test_mm512_maskz_dpwuud_epi32(__mmask16 __U, __m512i __A, __m512i __B, _
 
 __m512i test_mm512_dpwuuds_epi32(__m512i __A, __m512i __B, __m512i __C) {
 // CHECK-LABEL: @test_mm512_dpwuuds_epi32(
-// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwuuds.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwuuds.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
   return _mm512_dpwuuds_epi32(__A, __B, __C);
 }
 
 __m512i test_mm512_mask_dpwuuds_epi32(__m512i __A, __mmask16 __B, __m512i __C, __m512i __D) {
 // CHECK-LABEL: @test_mm512_mask_dpwuuds_epi32(
-// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwuuds.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwuuds.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
 // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
   return _mm512_mask_dpwuuds_epi32(__A, __B, __C, __D);
 }
 
 __m512i test_mm512_maskz_dpwuuds_epi32(__mmask16 __U, __m512i __A, __m512i __B, __m512i __C) {
 // CHECK-LABEL: @test_mm512_maskz_dpwuuds_epi32(
-// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwuuds.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+// CHECK: call <16 x i32> @llvm.x86.avx10.vpdpwuuds.512(<16 x i32> %{{.*}}, <32 x i16> %{{.*}}, <32 x i16> %{{.*}})
 // CHECK: zeroinitializer
 // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
   return _mm512_maskz_dpwuuds_epi32(__U, __A, __B, __C);
diff --git a/clang/test/CodeGen/X86/avx10_2ni-builtins.c b/clang/test/CodeGen/X86/avx10_2ni-builtins.c
index a250d91ae5989..be2719c33c52f 100644
--- a/clang/test/CodeGen/X86/avx10_2ni-builtins.c
+++ b/clang/test/CodeGen/X86/avx10_2ni-builtins.c
@@ -259,168 +259,168 @@ __m256i test_mm256_maskz_dpbuuds_epi32(__mmask8 __U, __m256i __W, __m256i __A, _
 // VNNI INT16
 __m128i test_mm_mask_dpwsud_epi32(__m128i __A, __mmask8 __B, __m128i __C, __m128i __D) {
 // CHECK-LABEL: @test_mm_mask_dpwsud_epi32(
-// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
 // CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
   return _mm_mask_dpwsud_epi32(__A, __B, __C, __D);
 }
 
 __m128i test_mm_maskz_dpwsud_epi32(__mmask8 __U, __m128i __A, __m128i __B, __m128i __C) {
 // CHECK-LABEL: @test_mm_maskz_dpwsud_epi32(
-// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
 // CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
   return _mm_maskz_dpwsud_epi32(__U, __A, __B, __C);
 }
 
 __m256i test_mm256_mask_dpwsud_epi32(__m256i __A, __mmask8 __B, __m256i __C, __m256i __D) {
 // CHECK-LABEL: @test_mm256_mask_dpwsud_epi32(
-// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
 // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_mask_dpwsud_epi32(__A, __B, __C, __D);
 }
 
 __m256i test_mm256_maskz_dpwsud_epi32(__mmask8 __U, __m256i __A, __m256i __B, __m256i __C) {
 // CHECK-LABEL: @test_mm256_maskz_dpwsud_epi32(
-// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
 // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_maskz_dpwsud_epi32(__U, __A, __B, __C);
 }
 
 __m128i test_mm_mask_dpwsuds_epi32(__m128i __A, __mmask8 __B, __m128i __C, __m128i __D) {
 // CHECK-LABEL: @test_mm_mask_dpwsuds_epi32(
-// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
 // CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
   return _mm_mask_dpwsuds_epi32(__A, __B, __C, __D);
 }
 
 __m128i test_mm_maskz_dpwsuds_epi32(__mmask8 __U, __m128i __A, __m128i __B, __m128i __C) {
 // CHECK-LABEL: @test_mm_maskz_dpwsuds_epi32(
-// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
 // CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
   return _mm_maskz_dpwsuds_epi32(__U, __A, __B, __C);
 }
 
 __m256i test_mm256_mask_dpwsuds_epi32(__m256i __A, __mmask8 __B, __m256i __C, __m256i __D) {
 // CHECK-LABEL: @test_mm256_mask_dpwsuds_epi32(
-// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
 // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_mask_dpwsuds_epi32(__A, __B, __C, __D);
 }
 
 __m256i test_mm256_maskz_dpwsuds_epi32(__mmask8 __U, __m256i __A, __m256i __B, __m256i __C) {
 // CHECK-LABEL: @test_mm256_maskz_dpwsuds_epi32(
-// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
 // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_maskz_dpwsuds_epi32(__U, __A, __B, __C);
 }
 
 __m128i test_mm_mask_dpwusd_epi32(__m128i __A, __mmask8 __B, __m128i __C, __m128i __D) {
 // CHECK-LABEL: @test_mm_mask_dpwusd_epi32(
-// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
 // CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
   return _mm_mask_dpwusd_epi32(__A, __B, __C, __D);
 }
 
 __m128i test_mm_maskz_dpwusd_epi32(__mmask8 __U, __m128i __A, __m128i __B, __m128i __C) {
 // CHECK-LABEL: @test_mm_maskz_dpwusd_epi32(
-// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
 // CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
   return _mm_maskz_dpwusd_epi32(__U, __A, __B, __C);
 }
 
 __m256i test_mm256_mask_dpwusd_epi32(__m256i __A, __mmask8 __B, __m256i __C, __m256i __D) {
 // CHECK-LABEL: @test_mm256_mask_dpwusd_epi32(
-// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
 // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_mask_dpwusd_epi32(__A, __B, __C, __D);
 }
 
 __m256i test_mm256_maskz_dpwusd_epi32(__mmask8 __U, __m256i __A, __m256i __B, __m256i __C) {
 // CHECK-LABEL: @test_mm256_maskz_dpwusd_epi32(
-// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
 // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_maskz_dpwusd_epi32(__U, __A, __B, __C);
 }
 
 __m128i test_mm_mask_dpwusds_epi32(__m128i __A, __mmask8 __B, __m128i __C, __m128i __D) {
 // CHECK-LABEL: @test_mm_mask_dpwusds_epi32(
-// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
 // CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
   return _mm_mask_dpwusds_epi32(__A, __B, __C, __D);
 }
 
 __m128i test_mm_maskz_dpwusds_epi32(__mmask8 __U, __m128i __A, __m128i __B, __m128i __C) {
 // CHECK-LABEL: @test_mm_maskz_dpwusds_epi32(
-// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
 // CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
   return _mm_maskz_dpwusds_epi32(__U, __A, __B, __C);
 }
 
 __m256i test_mm256_mask_dpwusds_epi32(__m256i __A, __mmask8 __B, __m256i __C, __m256i __D) {
 // CHECK-LABEL: @test_mm256_mask_dpwusds_epi32(
-// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
 // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_mask_dpwusds_epi32(__A, __B, __C, __D);
 }
 
 __m256i test_mm256_maskz_dpwusds_epi32(__mmask8 __U, __m256i __A, __m256i __B, __m256i __C) {
 // CHECK-LABEL: @test_mm256_maskz_dpwusds_epi32(
-// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
 // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_maskz_dpwusds_epi32(__U, __A, __B, __C);
 }
 
 __m128i test_mm_mask_dpwuud_epi32(__m128i __A, __mmask8 __B, __m128i __C, __m128i __D) {
 // CHECK-LABEL: @test_mm_mask_dpwuud_epi32(
-// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
 // CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
   return _mm_mask_dpwuud_epi32(__A, __B, __C, __D);
 }
 
 __m128i test_mm_maskz_dpwuud_epi32(__mmask8 __U, __m128i __A, __m128i __B, __m128i __C) {
 // CHECK-LABEL: @test_mm_maskz_dpwuud_epi32(
-// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
 // CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
   return _mm_maskz_dpwuud_epi32(__U, __A, __B, __C);
 }
 
 __m256i test_mm256_mask_dpwuud_epi32(__m256i __A, __mmask8 __B, __m256i __C, __m256i __D) {
 // CHECK-LABEL: @test_mm256_mask_dpwuud_epi32(
-// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
 // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_mask_dpwuud_epi32(__A, __B, __C, __D);
 }
 
 __m256i test_mm256_maskz_dpwuud_epi32(__mmask8 __U, __m256i __A, __m256i __B, __m256i __C) {
 // CHECK-LABEL: @test_mm256_maskz_dpwuud_epi32(
-// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
 // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_maskz_dpwuud_epi32(__U, __A, __B, __C);
 }
 
 __m128i test_mm_mask_dpwuuds_epi32(__m128i __A, __mmask8 __B, __m128i __C, __m128i __D) {
 // CHECK-LABEL: @test_mm_mask_dpwuuds_epi32(
-// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
 // CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
   return _mm_mask_dpwuuds_epi32(__A, __B, __C, __D);
 }
 
 __m128i test_mm_maskz_dpwuuds_epi32(__mmask8 __U, __m128i __A, __m128i __B, __m128i __C) {
 // CHECK-LABEL: @test_mm_maskz_dpwuuds_epi32(
-// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+// CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
 // CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
   return _mm_maskz_dpwuuds_epi32(__U, __A, __B, __C);
 }
 
 __m256i test_mm256_mask_dpwuuds_epi32(__m256i __A, __mmask8 __B, __m256i __C, __m256i __D) {
 // CHECK-LABEL: @test_mm256_mask_dpwuuds_epi32(
-// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
 // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_mask_dpwuuds_epi32(__A, __B, __C, __D);
 }
 
 __m256i test_mm256_maskz_dpwuuds_epi32(__mmask8 __U, __m256i __A, __m256i __B, __m256i __C) {
 // CHECK-LABEL: @test_mm256_maskz_dpwuuds_epi32(
-// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+// CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
 // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_maskz_dpwuuds_epi32(__U, __A, __B, __C);
 }
diff --git a/clang/test/CodeGen/X86/avxvnniint16-builtins.c b/clang/test/CodeGen/X86/avxvnniint16-builtins.c
index 941da9aa223b5..f28fff2c0cfec 100644
--- a/clang/test/CodeGen/X86/avxvnniint16-builtins.c
+++ b/clang/test/CodeGen/X86/avxvnniint16-builtins.c
@@ -11,72 +11,72 @@
 
 __m128i test_mm_dpwsud_epi32(__m128i __A, __m128i __B, __m128i __C) {
   // CHECK-LABEL: test_mm_dpwsud_epi32
-  // CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+  // CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
   return _mm_dpwsud_epi32(__A, __B, __C);
 }
 
 __m256i test_mm256_dpwsud_epi32(__m256i __A, __m256i __B, __m256i __C) {
   // CHECK-LABEL: test_mm256_dpwsud_epi32
-  // CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+  // CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
   return _mm256_dpwsud_epi32(__A, __B, __C);
 }
 
 __m128i test_mm_dpwsuds_epi32(__m128i __A, __m128i __B, __m128i __C) {
   // CHECK-LABEL: test_mm_dpwsuds_epi32
-  // CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+  // CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
   return _mm_dpwsuds_epi32(__A, __B, __C);
 }
 
 __m256i test_mm256_dpwsuds_epi32(__m256i __A, __m256i __B, __m256i __C) {
   // CHECK-LABEL: test_mm256_dpwsuds_epi32
-  // CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+  // CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
   return _mm256_dpwsuds_epi32(__A, __B, __C);
 }
 
 __m128i test_mm_dpwusd_epi32(__m128i __A, __m128i __B, __m128i __C) {
   // CHECK-LABEL: test_mm_dpwusd_epi32
-  // CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+  // CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
   return _mm_dpwusd_epi32(__A, __B, __C);
 }
 
 __m256i test_mm256_dpwusd_epi32(__m256i __A, __m256i __B, __m256i __C) {
   // CHECK-LABEL: test_mm256_dpwusd_epi32
-  // CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+  // CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
   return _mm256_dpwusd_epi32(__A, __B, __C);
 }
 
 __m128i test_mm_dpwusds_epi32(__m128i __A, __m128i __B, __m128i __C) {
   // CHECK-LABEL: test_mm_dpwusds_epi32
-  // CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+  // CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
   return _mm_dpwusds_epi32(__A, __B, __C);
 }
 
 __m256i test_mm256_dpwusds_epi32(__m256i __A, __m256i __B, __m256i __C) {
   // CHECK-LABEL: test_mm256_dpwusds_epi32
-  // CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+  // CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
   return _mm256_dpwusds_epi32(__A, __B, __C);
 }
 
 __m128i test_mm_dpwuud_epi32(__m128i __A, __m128i __B, __m128i __C) {
   // CHECK-LABEL: test_mm_dpwuud_epi32
-  // CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+  // CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
   return _mm_dpwuud_epi32(__A, __B, __C);
 }
 
 __m256i test_mm256_dpwuud_epi32(__m256i __A, __m256i __B, __m256i __C) {
   // CHECK-LABEL: test_mm256_dpwuud_epi32
-  // CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+  // CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
   return _mm256_dpwuud_epi32(__A, __B, __C);
 }
 
 __m128i test_mm_dpwuuds_epi32(__m128i __A, __m128i __B, __m128i __C) {
   // CHECK-LABEL: test_mm_dpwuuds_epi32
-  // CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
+  // CHECK: call <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> %{{.*}}, <8 x i16> %{{.*}}, <8 x i16> %{{.*}})
   return _mm_dpwuuds_epi32(__A, __B, __C);
 }
 
 __m256i test_mm256_dpwuuds_epi32(__m256i __A, __m256i __B, __m256i __C) {
   // CHECK-LABEL: test_mm256_dpwuuds_epi32
-  // CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
+  // CHECK: call <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> %{{.*}}, <16 x i16> %{{.*}}, <16 x i16> %{{.*}})
   return _mm256_dpwuuds_epi32(__A, __B, __C);
 }
diff --git a/llvm/include/llvm/IR/IntrinsicsX86.td b/llvm/include/llvm/IR/IntrinsicsX86.td
index 546eed7c904a1..ec80ba3e1ee81 100644
--- a/llvm/include/llvm/IR/IntrinsicsX86.td
+++ b/llvm/include/llvm/IR/IntrinsicsX86.td
@@ -1980,62 +1980,62 @@ let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
   def int_x86_avx2_vpdpwsud_128
       : ClangBuiltin<"__builtin_ia32_vpdpwsud128">,
         DefaultAttrsIntrinsic<[llvm_v4i32_ty],
-                              [llvm_v4i32_ty, llvm_v4i32_ty, llvm_v4i32_ty],
+                              [llvm_v4i32_ty, llvm_v8i16_ty, llvm_v8i16_ty],
                               [IntrNoMem]>;
   def int_x86_avx2_vpdpwsud_256
       : ClangBuiltin<"__builtin_ia32_vpdpwsud256">,
         DefaultAttrsIntrinsic<[llvm_v8i32_ty],
-                              [llvm_v8i32_ty, llvm_v8i32_ty, llvm_v8i32_ty],
+                              [llvm_v8i32_ty, llvm_v16i16_ty, llvm_v16i16_ty],
                               [IntrNoMem]>;
   def int_x86_avx2_vpdpwsuds_128
       : ClangBuiltin<"__builtin_ia32_vpdpwsuds128">,
         DefaultAttrsIntrinsic<[llvm_v4i32_ty],
-                              [llvm_v4i32_ty, llvm_v4i32_ty, llvm_v4i32_ty],
+                              [llvm_v4i32_ty, llvm_v8i16_ty, llvm_v8i16_ty],
                               [IntrNoMem]>;
   def int_x86_avx2_vpdpwsuds_256
       : ClangBuiltin<"__builtin_ia32_vpdpwsuds256">,
         DefaultAttrsIntrinsic<[llvm_v8i32_ty],
-                              [llvm_v8i32_ty, llvm_v8i32_ty, llvm_v8i32_ty],
+                              [llvm_v8i32_ty, llvm_v16i16_ty, llvm_v16i16_ty],
                               [IntrNoMem]>;
   def int_x86_avx2_vpdpwusd_128
       : ClangBuiltin<"__builtin_ia32_vpdpwusd128">,
         DefaultAttrsIntrinsic<[llvm_v4i32_ty],
-                              [llvm_v4i32_ty, llvm_v4i32_ty, llvm_v4i32_ty],
+                              [llvm_v4i32_ty, llvm_v8i16_ty, llvm_v8i16_ty],
                               [IntrNoMem]>;
   def int_x86_avx2_vpdpwusd_256
       : ClangBuiltin<"__builtin_ia32_vpdpwusd256">,
         DefaultAttrsIntrinsic<[llvm_v8i32_ty],
-                              [llvm_v8i32_ty, llvm_v8i32_ty, llvm_v8i32_ty],
+                              [llvm_v8i32_ty, llvm_v16i16_ty, llvm_v16i16_ty],
                               [IntrNoMem]>;
   def int_x86_avx2_vpdpwusds_128
       : ClangBuiltin<"__builtin_ia32_vpdpwusds128">,
         DefaultAttrsIntrinsic<[llvm_v4i32_ty],
-                              [llvm_v4i32_ty, llvm_v4i32_ty, llvm_v4i32_ty],
+                              [llvm_v4i32_ty, llvm_v8i16_ty, llvm_v8i16_ty],
                               [IntrNoMem]>;
   def int_x86_avx2_vpdpwusds_256
       : ClangBuiltin<"__builtin_ia32_vpdpwusds256">,
         DefaultAttrsIntrinsic<[llvm_v8i32_ty],
-                              [llvm_v8i32_ty, llvm_v8i32_ty, llvm_v8i32_ty],
+                              [llvm_v8i32_ty, llvm_v16i16_ty, llvm_v16i16_ty],
                               [IntrNoMem]>;
   def int_x86_avx2_vpdpwuud_128
       : ClangBuiltin<"__builtin_ia32_vpdpwuud128">,
         DefaultAttrsIntrinsic<[llvm_v4i32_ty],
-                              [llvm_v4i32_ty, llvm_v4i32_ty, llvm_v4i32_ty],
+                              [llvm_v4i32_ty, llvm_v8i16_ty, llvm_v8i16_ty],
                               [IntrNoMem]>;
   def int_x86_avx2_vpdpwuud_256
       : ClangBuiltin<"__builtin_ia32_vpdpwuud256">,
         DefaultAttrsIntrinsic<[llvm_v8i32_ty],
-                              [llvm_v8i32_ty, llvm_v8i32_ty, llvm_v8i32_ty],
+                              [llvm_v8i32_ty, llvm_v16i16_ty, llvm_v16i16_ty],
                               [IntrNoMem]>;
   def int_x86_avx2_vpdpwuuds_128
       : ClangBuiltin<"__builtin_ia32_vpdpwuuds128">,
         DefaultAttrsIntrinsic<[llvm_v4i32_ty],
-                              [llvm_v4i32_ty, llvm_v4i32_ty, llvm_v4i32_ty],
+                              [llvm_v4i32_ty, llvm_v8i16_ty, llvm_v8i16_ty],
                               [IntrNoMem]>;
   def int_x86_avx2_vpdpwuuds_256
       : ClangBuiltin<"__builtin_ia32_vpdpwuuds256">,
         DefaultAttrsIntrinsic<[llvm_v8i32_ty],
-                              [llvm_v8i32_ty, llvm_v8i32_ty, llvm_v8i32_ty],
+                              [llvm_v8i32_ty, llvm_v16i16_ty, llvm_v16i16_ty],
                               [IntrNoMem]>;
 }
 
@@ -5031,32 +5031,32 @@ let TargetPrefix = "x86" in {
   def int_x86_avx10_vpdpwsud_512 :
       ClangBuiltin<"__builtin_ia32_vpdpwsud512">,
       DefaultAttrsIntrinsic<[llvm_v16i32_ty],
-                            [llvm_v16i32_ty, llvm_v16i32_ty, llvm_v16i32_ty],
+                            [llvm_v16i32_ty, llvm_v32i16_ty, llvm_v32i16_ty],
                             [IntrNoMem]>;
   def int_x86_avx10_vpdpwsuds_512 :
       ClangBuiltin<"__builtin_ia32_vpdpwsuds512">,
       DefaultAttrsIntrinsic<[llvm_v16i32_ty],
-                            [llvm_v16i32_ty, llvm_v16i32_ty, llvm_v16i32_ty],
+                            [llvm_v16i32_ty, llvm_v32i16_ty, llvm_v32i16_ty],
                             [IntrNoMem]>;
   def int_x86_avx10_vpdpwusd_512 :
       ClangBuiltin<"__builtin_ia32_vpdpwusd512">,
       DefaultAttrsIntrinsic<[llvm_v16i32_ty],
-                            [llvm_v16i32_ty, llvm_v16i32_ty, llvm_v16i32_ty],
+                            [llvm_v16i32_ty, llvm_v32i16_ty, llvm_v32i16_ty],
                             [IntrNoMem]>;
   def int_x86_avx10_vpdpwusds_512 :
       ClangBuiltin<"__builtin_ia32_vpdpwusds512">,
       DefaultAttrsIntrinsic<[llvm_v16i32_ty],
-                            [llvm_v16i32_ty, llvm_v16i32_ty, llvm_v16i32_ty],
+                            [llvm_v16i32_ty, llvm_v32i16_ty, llvm_v32i16_ty],
                             [IntrNoMem]>;
   def int_x86_avx10_vpdpwuud_512 :
       ClangBuiltin<"__builtin_ia32_vpdpwuud512">,
       DefaultAttrsIntrinsic<[llvm_v16i32_ty],
-                            [llvm_v16i32_ty, llvm_v16i32_ty, llvm_v16i32_ty],
+                            [llvm_v16i32_ty, llvm_v32i16_ty, llvm_v32i16_ty],
                             [IntrNoMem]>;
   def int_x86_avx10_vpdpwuuds_512 :
       ClangBuiltin<"__builtin_ia32_vpdpwuuds512">,
       DefaultAttrsIntrinsic<[llvm_v16i32_ty],
-                            [llvm_v16i32_ty, llvm_v16i32_ty, llvm_v16i32_ty],
+                            [llvm_v16i32_ty, llvm_v32i16_ty, llvm_v32i16_ty],
                             [IntrNoMem]>;
 
   // VMPSADBW
diff --git a/llvm/lib/IR/AutoUpgrade.cpp b/llvm/lib/IR/AutoUpgrade.cpp
index 53ae18c4591f7..2ae65efc70da6 100644
--- a/llvm/lib/IR/AutoUpgrade.cpp
+++ b/llvm/lib/IR/AutoUpgrade.cpp
@@ -625,39 +625,72 @@ static bool upgradeX86IntrinsicFunction(Function *F, StringRef Name,
     return false; // No other 'x86.avx512.*'.
   }
 
-  if (Name.consume_front("avx2.vpdpb")) {
-    // Added in 21.1
-    ID = StringSwitch<Intrinsic::ID>(Name)
-             .Case("ssd.128", Intrinsic::x86_avx2_vpdpbssd_128)
-             .Case("ssd.256", Intrinsic::x86_avx2_vpdpbssd_256)
-             .Case("ssds.128", Intrinsic::x86_avx2_vpdpbssds_128)
-             .Case("ssds.256", Intrinsic::x86_avx2_vpdpbssds_256)
-             .Case("sud.128", Intrinsic::x86_avx2_vpdpbsud_128)
-             .Case("sud.256", Intrinsic::x86_avx2_vpdpbsud_256)
-             .Case("suds.128", Intrinsic::x86_avx2_vpdpbsuds_128)
-             .Case("suds.256", Intrinsic::x86_avx2_vpdpbsuds_256)
-             .Case("uud.128", Intrinsic::x86_avx2_vpdpbuud_128)
-             .Case("uud.256", Intrinsic::x86_avx2_vpdpbuud_256)
-             .Case("uuds.128", Intrinsic::x86_avx2_vpdpbuuds_128)
-             .Case("uuds.256", Intrinsic::x86_avx2_vpdpbuuds_256)
-             .Default(Intrinsic::not_intrinsic);
-    if (ID != Intrinsic::not_intrinsic)
-      return upgradeX86MultiplyAddBytes(F, ID, NewFn);
+  if (Name.consume_front("avx2.")) {
+    if (Name.consume_front("vpdpb")) {
+      // Added in 21.1
+      ID = StringSwitch<Intrinsic::ID>(Name)
+               .Case("ssd.128", Intrinsic::x86_avx2_vpdpbssd_128)
+               .Case("ssd.256", Intrinsic::x86_avx2_vpdpbssd_256)
+               .Case("ssds.128", Intrinsic::x86_avx2_vpdpbssds_128)
+               .Case("ssds.256", Intrinsic::x86_avx2_vpdpbssds_256)
+               .Case("sud.128", Intrinsic::x86_avx2_vpdpbsud_128)
+               .Case("sud.256", Intrinsic::x86_avx2_vpdpbsud_256)
+               .Case("suds.128", Intrinsic::x86_avx2_vpdpbsuds_128)
+               .Case("suds.256", Intrinsic::x86_avx2_vpdpbsuds_256)
+               .Case("uud.128", Intrinsic::x86_avx2_vpdpbuud_128)
+               .Case("uud.256", Intrinsic::x86_avx2_vpdpbuud_256)
+               .Case("uuds.128", Intrinsic::x86_avx2_vpdpbuuds_128)
+               .Case("uuds.256", Intrinsic::x86_avx2_vpdpbuuds_256)
+               .Default(Intrinsic::not_intrinsic);
+      if (ID != Intrinsic::not_intrinsic)
+        return upgradeX86MultiplyAddBytes(F, ID, NewFn);
+    } else if (Name.consume_front("vpdpw")) {
+      // Added in 21.1
+      ID = StringSwitch<Intrinsic::ID>(Name)
+               .Case("sud.128", Intrinsic::x86_avx2_vpdpwsud_128)
+               .Case("sud.256", Intrinsic::x86_avx2_vpdpwsud_256)
+               .Case("suds.128", Intrinsic::x86_avx2_vpdpwsuds_128)
+               .Case("suds.256", Intrinsic::x86_avx2_vpdpwsuds_256)
+               .Case("usd.128", Intrinsic::x86_avx2_vpdpwusd_128)
+               .Case("usd.256", Intrinsic::x86_avx2_vpdpwusd_256)
+               .Case("usds.128", Intrinsic::x86_avx2_vpdpwusds_128)
+               .Case("usds.256", Intrinsic::x86_avx2_vpdpwusds_256)
+               .Case("uud.128", Intrinsic::x86_avx2_vpdpwuud_128)
+               .Case("uud.256", Intrinsic::x86_avx2_vpdpwuud_256)
+               .Case("uuds.128", Intrinsic::x86_avx2_vpdpwuuds_128)
+               .Case("uuds.256", Intrinsic::x86_avx2_vpdpwuuds_256)
+               .Default(Intrinsic::not_intrinsic);
+      if (ID != Intrinsic::not_intrinsic)
+        return upgradeX86MultiplyAddWords(F, ID, NewFn);
+    }
     return false; // No other 'x86.avx2.*'
   }
 
-  if (Name.consume_front("avx10.vpdpb")) {
-    // Added in 21.1
-    ID = StringSwitch<Intrinsic::ID>(Name)
-             .Case("ssd.512", Intrinsic::x86_avx10_vpdpbssd_512)
-             .Case("ssds.512", Intrinsic::x86_avx10_vpdpbssds_512)
-             .Case("sud.512", Intrinsic::x86_avx10_vpdpbsud_512)
-             .Case("suds.512", Intrinsic::x86_avx10_vpdpbsuds_512)
-             .Case("uud.512", Intrinsic::x86_avx10_vpdpbuud_512)
-             .Case("uuds.512", Intrinsic::x86_avx10_vpdpbuuds_512)
-             .Default(Intrinsic::not_intrinsic);
-    if (ID != Intrinsic::not_intrinsic)
-      return upgradeX86MultiplyAddBytes(F, ID, NewFn);
+  if (Name.consume_front("avx10.")) {
+    if (Name.consume_front("vpdpb")) {
+      // Added in 21.1
+      ID = StringSwitch<Intrinsic::ID>(Name)
+               .Case("ssd.512", Intrinsic::x86_avx10_vpdpbssd_512)
+               .Case("ssds.512", Intrinsic::x86_avx10_vpdpbssds_512)
+               .Case("sud.512", Intrinsic::x86_avx10_vpdpbsud_512)
+               .Case("suds.512", Intrinsic::x86_avx10_vpdpbsuds_512)
+               .Case("uud.512", Intrinsic::x86_avx10_vpdpbuud_512)
+               .Case("uuds.512", Intrinsic::x86_avx10_vpdpbuuds_512)
+               .Default(Intrinsic::not_intrinsic);
+      if (ID != Intrinsic::not_intrinsic)
+        return upgradeX86MultiplyAddBytes(F, ID, NewFn);
+    } else if (Name.consume_front("vpdpw")) {
+      ID = StringSwitch<Intrinsic::ID>(Name)
+               .Case("sud.512", Intrinsic::x86_avx10_vpdpwsud_512)
+               .Case("suds.512", Intrinsic::x86_avx10_vpdpwsuds_512)
+               .Case("usd.512", Intrinsic::x86_avx10_vpdpwusd_512)
+               .Case("usds.512", Intrinsic::x86_avx10_vpdpwusds_512)
+               .Case("uud.512", Intrinsic::x86_avx10_vpdpwuud_512)
+               .Case("uuds.512", Intrinsic::x86_avx10_vpdpwuuds_512)
+               .Default(Intrinsic::not_intrinsic);
+      if (ID != Intrinsic::not_intrinsic)
+        return upgradeX86MultiplyAddWords(F, ID, NewFn);
+    }
     return false; // No other 'x86.avx10.*'
   }
 
@@ -5445,6 +5478,24 @@ void llvm::UpgradeIntrinsicCall(CallBase *CI, Function *NewFn) {
   case Intrinsic::x86_avx512_vpdpwssds_128:
   case Intrinsic::x86_avx512_vpdpwssds_256:
   case Intrinsic::x86_avx512_vpdpwssds_512:
+  case Intrinsic::x86_avx2_vpdpwsud_128:
+  case Intrinsic::x86_avx2_vpdpwsud_256:
+  case Intrinsic::x86_avx10_vpdpwsud_512:
+  case Intrinsic::x86_avx2_vpdpwsuds_128:
+  case Intrinsic::x86_avx2_vpdpwsuds_256:
+  case Intrinsic::x86_avx10_vpdpwsuds_512:
+  case Intrinsic::x86_avx2_vpdpwusd_128:
+  case Intrinsic::x86_avx2_vpdpwusd_256:
+  case Intrinsic::x86_avx10_vpdpwusd_512:
+  case Intrinsic::x86_avx2_vpdpwusds_128:
+  case Intrinsic::x86_avx2_vpdpwusds_256:
+  case Intrinsic::x86_avx10_vpdpwusds_512:
+  case Intrinsic::x86_avx2_vpdpwuud_128:
+  case Intrinsic::x86_avx2_vpdpwuud_256:
+  case Intrinsic::x86_avx10_vpdpwuud_512:
+  case Intrinsic::x86_avx2_vpdpwuuds_128:
+  case Intrinsic::x86_avx2_vpdpwuuds_256:
+  case Intrinsic::x86_avx10_vpdpwuuds_512:
     unsigned NumElts = CI->getType()->getPrimitiveSizeInBits() / 16;
     Value *Args[] = {CI->getArgOperand(0), CI->getArgOperand(1),
                      CI->getArgOperand(2)};
diff --git a/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp b/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp
index f0fe71e9eb69e..54d5145ad64a5 100644
--- a/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp
+++ b/llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp
@@ -5889,6 +5889,54 @@ struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> {
     //   <16 x i32> @llvm.x86.avx512.vpdpwssds.512
     //                  (<16 x i32>, <32 x i16>, <32 x i16>)
     //
+    // Multiply and Add Signed and Unsigned Word Integers
+    //   < 4 x i32> @llvm.x86.avx2.vpdpwsud.128
+    //                  (< 4 x i32>, < 8 x i16>, < 8 x i16>)
+    //   < 8 x i32> @llvm.x86.avx2.vpdpwsud.256
+    //                  (< 8 x i32>, <16 x i16>, <16 x i16>)
+    //   <16 x i32> @llvm.x86.avx10.vpdpwsud.512
+    //                  (<16 x i32>, <32 x i16>, <32 x i16>)
+    //
+    // Multiply and Add Signed and Unsigned Word Integers With Saturation
+    //   < 4 x i32> @llvm.x86.avx2.vpdpwsuds.128
+    //                  (< 4 x i32>, < 8 x i16>, < 8 x i16>)
+    //   < 8 x i32> @llvm.x86.avx2.vpdpwsuds.256
+    //                  (< 8 x i32>, <16 x i16>, <16 x i16>)
+    //   <16 x i32> @llvm.x86.avx10.vpdpwsuds.512
+    //                  (<16 x i32>, <32 x i16>, <32 x i16>)
+    //
+    // Multiply and Add Unsigned and Signed Word Integers
+    //   < 4 x i32> @llvm.x86.avx2.vpdpwusd.128
+    //                  (< 4 x i32>, < 8 x i16>, < 8 x i16>)
+    //   < 8 x i32> @llvm.x86.avx2.vpdpwusd.256
+    //                  (< 8 x i32>, <16 x i16>, <16 x i16>)
+    //   <16 x i32> @llvm.x86.avx10.vpdpwusd.512
+    //                  (<16 x i32>, <32 x i16>, <32 x i16>)
+    //
+    // Multiply and Add Unsigned and Signed Word Integers With Saturation
+    //   < 4 x i32> @llvm.x86.avx2.vpdpwusds.128
+    //                  (< 4 x i32>, < 8 x i16>, < 8 x i16>)
+    //   < 8 x i32> @llvm.x86.avx2.vpdpwusds.256
+    //                  (< 8 x i32>, <16 x i16>, <16 x i16>)
+    //   <16 x i32> @llvm.x86.avx10.vpdpwusds.512
+    //                  (<16 x i32>, <32 x i16>, <32 x i16>)
+    //
+    // Multiply and Add Unsigned and Unsigned Word Integers
+    //   < 4 x i32> @llvm.x86.avx2.vpdpwuud.128
+    //                  (< 4 x i32>, < 8 x i16>, < 8 x i16>)
+    //   < 8 x i32> @llvm.x86.avx2.vpdpwuud.256
+    //                  (< 8 x i32>, <16 x i16>, <16 x i16>)
+    //   <16 x i32> @llvm.x86.avx10.vpdpwuud.512
+    //                  (<16 x i32>, <32 x i16>, <32 x i16>)
+    //
+    // Multiply and Add Unsigned and Unsigned Word Integers With Saturation
+    //   < 4 x i32> @llvm.x86.avx2.vpdpwuuds.128
+    //                  (< 4 x i32>, < 8 x i16>, < 8 x i16>)
+    //   < 8 x i32> @llvm.x86.avx2.vpdpwuuds.256
+    //                  (< 8 x i32>, <16 x i16>, <16 x i16>)
+    //   <16 x i32> @llvm.x86.avx10.vpdpwuuds.512
+    //                  (<16 x i32>, <32 x i16>, <32 x i16>)
+    //
     // These intrinsics are auto-upgraded into non-masked forms:
     //   <4 x i32> @llvm.x86.avx512.mask.vpdpwssd.128
     //                 (<4 x i32>, <8 x i16>, <8 x i16>, i8)
@@ -5921,6 +5969,24 @@ struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> {
     case Intrinsic::x86_avx512_vpdpwssds_128:
     case Intrinsic::x86_avx512_vpdpwssds_256:
     case Intrinsic::x86_avx512_vpdpwssds_512:
+    case Intrinsic::x86_avx2_vpdpwsud_128:
+    case Intrinsic::x86_avx2_vpdpwsud_256:
+    case Intrinsic::x86_avx10_vpdpwsud_512:
+    case Intrinsic::x86_avx2_vpdpwsuds_128:
+    case Intrinsic::x86_avx2_vpdpwsuds_256:
+    case Intrinsic::x86_avx10_vpdpwsuds_512:
+    case Intrinsic::x86_avx2_vpdpwusd_128:
+    case Intrinsic::x86_avx2_vpdpwusd_256:
+    case Intrinsic::x86_avx10_vpdpwusd_512:
+    case Intrinsic::x86_avx2_vpdpwusds_128:
+    case Intrinsic::x86_avx2_vpdpwusds_256:
+    case Intrinsic::x86_avx10_vpdpwusds_512:
+    case Intrinsic::x86_avx2_vpdpwuud_128:
+    case Intrinsic::x86_avx2_vpdpwuud_256:
+    case Intrinsic::x86_avx10_vpdpwuud_512:
+    case Intrinsic::x86_avx2_vpdpwuuds_128:
+    case Intrinsic::x86_avx2_vpdpwuuds_256:
+    case Intrinsic::x86_avx10_vpdpwuuds_512:
       handleVectorPmaddIntrinsic(I, /*ReductionFactor=*/2,
                                  /*ZeroPurifies=*/true, /*EltSizeInBits=*/16);
       break;
diff --git a/llvm/test/CodeGen/X86/avx10.2-intrinsic-upgrade.ll b/llvm/test/CodeGen/X86/avx10.2-intrinsic-upgrade.ll
index 76d84c1159ee4..860d60ff0d4e1 100644
--- a/llvm/test/CodeGen/X86/avx10.2-intrinsic-upgrade.ll
+++ b/llvm/test/CodeGen/X86/avx10.2-intrinsic-upgrade.ll
@@ -97,3 +97,99 @@ define <16 x i32>@test_int_x86_avx10_vpdpbuuds_512(<16 x i32> %x0, <16 x i32> %x
   %res = call <16 x i32> @llvm.x86.avx10.vpdpbuuds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
   ret <16 x i32> %res
 }
+
+declare <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32>, <16 x i32>, <16 x i32>)
+
+define <16 x i32>@test_int_x86_avx10_vpdpwsud_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
+; X86-LABEL: test_int_x86_avx10_vpdpwsud_512:
+; X86:       # %bb.0:
+; X86-NEXT:    vpdpwsud %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x76,0x48,0xd2,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx10_vpdpwsud_512:
+; X64:       # %bb.0:
+; X64-NEXT:    vpdpwsud %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x76,0x48,0xd2,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
+  ret <16 x i32> %res
+}
+
+declare <16 x i32> @llvm.x86.avx10.vpdpwsuds.512(<16 x i32>, <16 x i32>, <16 x i32>)
+
+define <16 x i32>@test_int_x86_avx10_vpdpwsuds_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
+; X86-LABEL: test_int_x86_avx10_vpdpwsuds_512:
+; X86:       # %bb.0:
+; X86-NEXT:    vpdpwsuds %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x76,0x48,0xd3,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx10_vpdpwsuds_512:
+; X64:       # %bb.0:
+; X64-NEXT:    vpdpwsuds %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x76,0x48,0xd3,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <16 x i32> @llvm.x86.avx10.vpdpwsuds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
+  ret <16 x i32> %res
+}
+
+declare <16 x i32> @llvm.x86.avx10.vpdpwusd.512(<16 x i32>, <16 x i32>, <16 x i32>)
+
+define <16 x i32>@test_int_x86_avx10_vpdpwusd_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
+; X86-LABEL: test_int_x86_avx10_vpdpwusd_512:
+; X86:       # %bb.0:
+; X86-NEXT:    vpdpwusd %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0xd2,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx10_vpdpwusd_512:
+; X64:       # %bb.0:
+; X64-NEXT:    vpdpwusd %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0xd2,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <16 x i32> @llvm.x86.avx10.vpdpwusd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
+  ret <16 x i32> %res
+}
+
+declare <16 x i32> @llvm.x86.avx10.vpdpwusds.512(<16 x i32>, <16 x i32>, <16 x i32>)
+
+define <16 x i32>@test_int_x86_avx10_vpdpwusds_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
+; X86-LABEL: test_int_x86_avx10_vpdpwusds_512:
+; X86:       # %bb.0:
+; X86-NEXT:    vpdpwusds %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0xd3,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx10_vpdpwusds_512:
+; X64:       # %bb.0:
+; X64-NEXT:    vpdpwusds %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0xd3,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <16 x i32> @llvm.x86.avx10.vpdpwusds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
+  ret <16 x i32> %res
+}
+
+declare <16 x i32> @llvm.x86.avx10.vpdpwuud.512(<16 x i32>, <16 x i32>, <16 x i32>)
+
+define <16 x i32>@test_int_x86_avx10_vpdpwuud_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
+; X86-LABEL: test_int_x86_avx10_vpdpwuud_512:
+; X86:       # %bb.0:
+; X86-NEXT:    vpdpwuud %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x74,0x48,0xd2,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx10_vpdpwuud_512:
+; X64:       # %bb.0:
+; X64-NEXT:    vpdpwuud %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x74,0x48,0xd2,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <16 x i32> @llvm.x86.avx10.vpdpwuud.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
+  ret <16 x i32> %res
+}
+
+declare <16 x i32> @llvm.x86.avx10.vpdpwuuds.512(<16 x i32>, <16 x i32>, <16 x i32>)
+
+define <16 x i32>@test_int_x86_avx10_vpdpwuuds_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
+; X86-LABEL: test_int_x86_avx10_vpdpwuuds_512:
+; X86:       # %bb.0:
+; X86-NEXT:    vpdpwuuds %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x74,0x48,0xd3,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx10_vpdpwuuds_512:
+; X64:       # %bb.0:
+; X64-NEXT:    vpdpwuuds %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x74,0x48,0xd3,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <16 x i32> @llvm.x86.avx10.vpdpwuuds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2)
+  ret <16 x i32> %res
+}
diff --git a/llvm/test/CodeGen/X86/avx10_2_512ni-intrinsics.ll b/llvm/test/CodeGen/X86/avx10_2_512ni-intrinsics.ll
index a2aad604f19bc..e9c6cb6a19ba4 100644
--- a/llvm/test/CodeGen/X86/avx10_2_512ni-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx10_2_512ni-intrinsics.ll
@@ -220,7 +220,7 @@ declare <16 x i32> @llvm.x86.avx10.vpdpbuuds.512(<16 x i32>, <64 x i8>, <64 x i8
 
 ; VNNI INT16
 
-define <16 x i32> @test_mm512_dpwsud_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr %pB) {
+define <16 x i32> @test_mm512_dpwsud_epi32(<16 x i32> %__W, <32 x i16> %__A, ptr %pB) {
 ; X86-LABEL: test_mm512_dpwsud_epi32:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
@@ -231,12 +231,12 @@ define <16 x i32> @test_mm512_dpwsud_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr
 ; X64:       # %bb.0:
 ; X64-NEXT:    vpdpwsud (%rdi), %zmm1, %zmm0 # encoding: [0x62,0xf2,0x76,0x48,0xd2,0x07]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %__B = load <16 x i32>, ptr %pB
-  %res = tail call <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B)
+  %__B = load <32 x i16>, ptr %pB
+  %res = tail call <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32> %__W, <32 x i16> %__A, <32 x i16> %__B)
   ret <16 x i32> %res
 }
 
-define <16 x i32> @test_mm512_mask_dpwsuds_epi32(<16 x i32> %__W, i16 zeroext %__U, <16 x i32> %__A, <16 x i32> %__B) {
+define <16 x i32> @test_mm512_mask_dpwsuds_epi32(<16 x i32> %__W, i16 zeroext %__U, <32 x i16> %__A, <32 x i16> %__B) {
 ; X86-LABEL: test_mm512_mask_dpwsuds_epi32:
 ; X86:       # %bb.0:
 ; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
@@ -248,13 +248,13 @@ define <16 x i32> @test_mm512_mask_dpwsuds_epi32(<16 x i32> %__W, i16 zeroext %_
 ; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
 ; X64-NEXT:    vpdpwsuds %zmm2, %zmm1, %zmm0 {%k1} # encoding: [0x62,0xf2,0x76,0x49,0xd3,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpwsuds.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B)
+  %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpwsuds.512(<16 x i32> %__W, <32 x i16> %__A, <32 x i16> %__B)
   %bst = bitcast i16 %__U to <16 x i1>
   %res = select <16 x i1> %bst, <16 x i32> %dpi, <16 x i32> %__W
   ret <16 x i32> %res
 }
 
-define <16 x i32> @test_mm512_maskz_dpwsud_epi32(i16 zeroext %__U, <16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) {
+define <16 x i32> @test_mm512_maskz_dpwsud_epi32(i16 zeroext %__U, <16 x i32> %__W, <32 x i16> %__A, <32 x i16> %__B) {
 ; X86-LABEL: test_mm512_maskz_dpwsud_epi32:
 ; X86:       # %bb.0:
 ; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
@@ -266,14 +266,14 @@ define <16 x i32> @test_mm512_maskz_dpwsud_epi32(i16 zeroext %__U, <16 x i32> %_
 ; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
 ; X64-NEXT:    vpdpwsud %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x76,0xc9,0xd2,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B)
+  %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32> %__W, <32 x i16> %__A, <32 x i16> %__B)
   %bst = bitcast i16 %__U to <16 x i1>
   %res = select <16 x i1> %bst, <16 x i32> %dpi, <16 x i32> zeroinitializer
   ret <16 x i32> %res
 }
 
-declare <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32>, <16 x i32>, <16 x i32>)
-declare <16 x i32> @llvm.x86.avx10.vpdpwsuds.512(<16 x i32>, <16 x i32>, <16 x i32>)
+declare <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32>, <32 x i16>, <32 x i16>)
+declare <16 x i32> @llvm.x86.avx10.vpdpwsuds.512(<16 x i32>, <32 x i16>, <32 x i16>)
 
 define <16 x i32> @test_mm512_dpwusd_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr %pB) {
 ; X86-LABEL: test_mm512_dpwusd_epi32:
diff --git a/llvm/test/CodeGen/X86/avx10_2ni-intrinsics.ll b/llvm/test/CodeGen/X86/avx10_2ni-intrinsics.ll
index 1f270d539cdb4..bf7f9375570f9 100644
--- a/llvm/test/CodeGen/X86/avx10_2ni-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx10_2ni-intrinsics.ll
@@ -334,7 +334,7 @@ declare <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32>, <32 x i8>, <32 x i8>)
 
 ; VNNI INT16
 
-define <4 x i32> @test_mm_mask_dpwsud_epi32(<4 x i32> %__W, i4 zeroext %__U, <4 x i32> %__A, <4 x i32> %__B) {
+define <4 x i32> @test_mm_mask_dpwsud_epi32(<4 x i32> %__W, i4 zeroext %__U, <8 x i16> %__A, <8 x i16> %__B) {
 ; X86-LABEL: test_mm_mask_dpwsud_epi32:
 ; X86:       # %bb.0:
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
@@ -346,13 +346,13 @@ define <4 x i32> @test_mm_mask_dpwsud_epi32(<4 x i32> %__W, i4 zeroext %__U, <4
 ; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
 ; X64-NEXT:    vpdpwsud %xmm2, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0x76,0x09,0xd2,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B)
+  %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %__W, <8 x i16> %__A, <8 x i16> %__B)
   %bst = bitcast i4 %__U to <4 x i1>
   %res = select <4 x i1> %bst, <4 x i32> %dpi, <4 x i32> %__W
   ret <4 x i32> %res
 }
 
-define <4 x i32> @test_mm_maskz_dpwsuds_epi32(i4 zeroext %__U, <4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B) {
+define <4 x i32> @test_mm_maskz_dpwsuds_epi32(i4 zeroext %__U, <4 x i32> %__W, <8 x i16> %__A, <8 x i16> %__B) {
 ; X86-LABEL: test_mm_maskz_dpwsuds_epi32:
 ; X86:       # %bb.0:
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
@@ -364,13 +364,13 @@ define <4 x i32> @test_mm_maskz_dpwsuds_epi32(i4 zeroext %__U, <4 x i32> %__W, <
 ; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
 ; X64-NEXT:    vpdpwsuds %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x76,0x89,0xd3,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B)
+  %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %__W, <8 x i16> %__A, <8 x i16> %__B)
   %bst = bitcast i4 %__U to <4 x i1>
   %res = select <4 x i1> %bst, <4 x i32> %dpi, <4 x i32> zeroinitializer
   ret <4 x i32> %res
 }
 
-define <8 x i32> @test_mm256_maskz_dpwsuds_epi32(<8 x i32> %__W, i8 zeroext %__U, <8 x i32> %__A, <8 x i32> %__B) {
+define <8 x i32> @test_mm256_maskz_dpwsuds_epi32(<8 x i32> %__W, i8 zeroext %__U, <16 x i16> %__A, <16 x i16> %__B) {
 ; X86-LABEL: test_mm256_maskz_dpwsuds_epi32:
 ; X86:       # %bb.0:
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
@@ -382,13 +382,13 @@ define <8 x i32> @test_mm256_maskz_dpwsuds_epi32(<8 x i32> %__W, i8 zeroext %__U
 ; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
 ; X64-NEXT:    vpdpwsuds %ymm2, %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf2,0x76,0x29,0xd3,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B)
+  %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %__W, <16 x i16> %__A, <16 x i16> %__B)
   %bst = bitcast i8 %__U to <8 x i1>
   %res = select <8 x i1> %bst, <8 x i32> %dpi, <8 x i32> %__W
   ret <8 x i32> %res
 }
 
-define <8 x i32> @test_mm256_mask_dpwsud_epi32(i8 zeroext %__U, <8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B) {
+define <8 x i32> @test_mm256_mask_dpwsud_epi32(i8 zeroext %__U, <8 x i32> %__W, <16 x i16> %__A, <16 x i16> %__B) {
 ; X86-LABEL: test_mm256_mask_dpwsud_epi32:
 ; X86:       # %bb.0:
 ; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
@@ -400,16 +400,16 @@ define <8 x i32> @test_mm256_mask_dpwsud_epi32(i8 zeroext %__U, <8 x i32> %__W,
 ; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
 ; X64-NEXT:    vpdpwsud %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x76,0xa9,0xd2,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B)
+  %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %__W, <16 x i16> %__A, <16 x i16> %__B)
   %bst = bitcast i8 %__U to <8 x i1>
   %res = select <8 x i1> %bst, <8 x i32> %dpi, <8 x i32> zeroinitializer
   ret <8 x i32> %res
 }
 
-declare <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32>, <4 x i32>, <4 x i32>)
-declare <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32>, <4 x i32>, <4 x i32>)
-declare <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32>, <8 x i32>, <8 x i32>)
-declare <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32>, <8 x i32>, <8 x i32>)
+declare <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32>, <8 x i16>, <8 x i16>)
+declare <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32>, <8 x i16>, <8 x i16>)
+declare <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32>, <16 x i16>, <16 x i16>)
+declare <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32>, <16 x i16>, <16 x i16>)
 
 define <4 x i32> @test_mm_mask_dpwusd_epi32(<4 x i32> %__W, i4 zeroext %__U, <4 x i32> %__A, <4 x i32> %__B) {
 ; X86-LABEL: test_mm_mask_dpwusd_epi32:
diff --git a/llvm/test/CodeGen/X86/avxvnniint16-intrinsics-upgrade.ll b/llvm/test/CodeGen/X86/avxvnniint16-intrinsics-upgrade.ll
new file mode 100644
index 0000000000000..abdc296ae1e1c
--- /dev/null
+++ b/llvm/test/CodeGen/X86/avxvnniint16-intrinsics-upgrade.ll
@@ -0,0 +1,185 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -verify-machineinstrs -mtriple=x86_64-unknown-unknown --show-mc-encoding -mattr=+avxvnniint16 | FileCheck %s
+; RUN: llc < %s -verify-machineinstrs -mtriple=i686-unknown-unknown --show-mc-encoding -mattr=+avxvnniint16 | FileCheck %s
+; RUN: llc < %s -verify-machineinstrs -mtriple=x86_64-unknown-unknown --show-mc-encoding -mattr=+avx10.2 | FileCheck %s --check-prefix=AVX10
+; RUN: llc < %s -verify-machineinstrs -mtriple=i686-unknown-unknown --show-mc-encoding -mattr=+avx10.2 | FileCheck %s --check-prefix=AVX10
+
+define <4 x i32> @test_int_x86_avx2_vpdpwsud_128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C) {
+; CHECK-LABEL: test_int_x86_avx2_vpdpwsud_128:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpdpwsud %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x72,0xd2,0xc2]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+;
+; AVX10-LABEL: test_int_x86_avx2_vpdpwsud_128:
+; AVX10:       # %bb.0:
+; AVX10-NEXT:    vpdpwsud %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x76,0x08,0xd2,0xc2]
+; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+  ret <4 x i32> %ret
+}
+declare <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+
+define <8 x i32> @test_int_x86_avx2_vpdpwsud_256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C) {
+; CHECK-LABEL: test_int_x86_avx2_vpdpwsud_256:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpdpwsud %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x76,0xd2,0xc2]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+;
+; AVX10-LABEL: test_int_x86_avx2_vpdpwsud_256:
+; AVX10:       # %bb.0:
+; AVX10-NEXT:    vpdpwsud %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x76,0x28,0xd2,0xc2]
+; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+  ret <8 x i32> %ret
+}
+declare <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+
+define <4 x i32> @test_int_x86_avx2_vpdpwsuds_128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C) {
+; CHECK-LABEL: test_int_x86_avx2_vpdpwsuds_128:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpdpwsuds %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x72,0xd3,0xc2]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+;
+; AVX10-LABEL: test_int_x86_avx2_vpdpwsuds_128:
+; AVX10:       # %bb.0:
+; AVX10-NEXT:    vpdpwsuds %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x76,0x08,0xd3,0xc2]
+; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+  ret <4 x i32> %ret
+}
+declare <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+
+define <8 x i32> @test_int_x86_avx2_vpdpwsuds_256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C) {
+; CHECK-LABEL: test_int_x86_avx2_vpdpwsuds_256:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpdpwsuds %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x76,0xd3,0xc2]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+;
+; AVX10-LABEL: test_int_x86_avx2_vpdpwsuds_256:
+; AVX10:       # %bb.0:
+; AVX10-NEXT:    vpdpwsuds %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x76,0x28,0xd3,0xc2]
+; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+  ret <8 x i32> %ret
+}
+declare <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+
+define <4 x i32> @test_int_x86_avx2_vpdpwusd_128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C) {
+; CHECK-LABEL: test_int_x86_avx2_vpdpwusd_128:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpdpwusd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xd2,0xc2]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+;
+; AVX10-LABEL: test_int_x86_avx2_vpdpwusd_128:
+; AVX10:       # %bb.0:
+; AVX10-NEXT:    vpdpwusd %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x75,0x08,0xd2,0xc2]
+; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+  ret <4 x i32> %ret
+}
+declare <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+
+define <8 x i32> @test_int_x86_avx2_vpdpwusd_256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C) {
+; CHECK-LABEL: test_int_x86_avx2_vpdpwusd_256:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpdpwusd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0xd2,0xc2]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+;
+; AVX10-LABEL: test_int_x86_avx2_vpdpwusd_256:
+; AVX10:       # %bb.0:
+; AVX10-NEXT:    vpdpwusd %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x75,0x28,0xd2,0xc2]
+; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+  ret <8 x i32> %ret
+}
+declare <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+
+define <4 x i32> @test_int_x86_avx2_vpdpwusds_128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C) {
+; CHECK-LABEL: test_int_x86_avx2_vpdpwusds_128:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpdpwusds %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xd3,0xc2]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+;
+; AVX10-LABEL: test_int_x86_avx2_vpdpwusds_128:
+; AVX10:       # %bb.0:
+; AVX10-NEXT:    vpdpwusds %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x75,0x08,0xd3,0xc2]
+; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+  ret <4 x i32> %ret
+}
+declare <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+
+define <8 x i32> @test_int_x86_avx2_vpdpwusds_256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C) {
+; CHECK-LABEL: test_int_x86_avx2_vpdpwusds_256:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpdpwusds %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0xd3,0xc2]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+;
+; AVX10-LABEL: test_int_x86_avx2_vpdpwusds_256:
+; AVX10:       # %bb.0:
+; AVX10-NEXT:    vpdpwusds %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x75,0x28,0xd3,0xc2]
+; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+  ret <8 x i32> %ret
+}
+declare <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+
+define <4 x i32> @test_int_x86_avx2_vpdpwuud_128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C) {
+; CHECK-LABEL: test_int_x86_avx2_vpdpwuud_128:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpdpwuud %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x70,0xd2,0xc2]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+;
+; AVX10-LABEL: test_int_x86_avx2_vpdpwuud_128:
+; AVX10:       # %bb.0:
+; AVX10-NEXT:    vpdpwuud %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x74,0x08,0xd2,0xc2]
+; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+  ret <4 x i32> %ret
+}
+declare <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+
+define <8 x i32> @test_int_x86_avx2_vpdpwuud_256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C) {
+; CHECK-LABEL: test_int_x86_avx2_vpdpwuud_256:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpdpwuud %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x74,0xd2,0xc2]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+;
+; AVX10-LABEL: test_int_x86_avx2_vpdpwuud_256:
+; AVX10:       # %bb.0:
+; AVX10-NEXT:    vpdpwuud %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x74,0x28,0xd2,0xc2]
+; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+  ret <8 x i32> %ret
+}
+declare <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+
+define <4 x i32> @test_int_x86_avx2_vpdpwuuds_128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C) {
+; CHECK-LABEL: test_int_x86_avx2_vpdpwuuds_128:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpdpwuuds %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x70,0xd3,0xc2]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+;
+; AVX10-LABEL: test_int_x86_avx2_vpdpwuuds_128:
+; AVX10:       # %bb.0:
+; AVX10-NEXT:    vpdpwuuds %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x74,0x08,0xd3,0xc2]
+; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+  ret <4 x i32> %ret
+}
+declare <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+
+define <8 x i32> @test_int_x86_avx2_vpdpwuuds_256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C) {
+; CHECK-LABEL: test_int_x86_avx2_vpdpwuuds_256:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpdpwuuds %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x74,0xd3,0xc2]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+;
+; AVX10-LABEL: test_int_x86_avx2_vpdpwuuds_256:
+; AVX10:       # %bb.0:
+; AVX10-NEXT:    vpdpwuuds %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x74,0x28,0xd3,0xc2]
+; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+  ret <8 x i32> %ret
+}
+declare <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
diff --git a/llvm/test/CodeGen/X86/avxvnniint16-intrinsics.ll b/llvm/test/CodeGen/X86/avxvnniint16-intrinsics.ll
index abdc296ae1e1c..7576b12645bd0 100644
--- a/llvm/test/CodeGen/X86/avxvnniint16-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avxvnniint16-intrinsics.ll
@@ -4,7 +4,7 @@
 ; RUN: llc < %s -verify-machineinstrs -mtriple=x86_64-unknown-unknown --show-mc-encoding -mattr=+avx10.2 | FileCheck %s --check-prefix=AVX10
 ; RUN: llc < %s -verify-machineinstrs -mtriple=i686-unknown-unknown --show-mc-encoding -mattr=+avx10.2 | FileCheck %s --check-prefix=AVX10
 
-define <4 x i32> @test_int_x86_avx2_vpdpwsud_128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C) {
+define <4 x i32> @test_int_x86_avx2_vpdpwsud_128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C) {
 ; CHECK-LABEL: test_int_x86_avx2_vpdpwsud_128:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpdpwsud %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x72,0xd2,0xc2]
@@ -14,12 +14,12 @@ define <4 x i32> @test_int_x86_avx2_vpdpwsud_128(<4 x i32> %A, <4 x i32> %B, <4
 ; AVX10:       # %bb.0:
 ; AVX10-NEXT:    vpdpwsud %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x76,0x08,0xd2,0xc2]
 ; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
   ret <4 x i32> %ret
 }
-declare <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+declare <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
 
-define <8 x i32> @test_int_x86_avx2_vpdpwsud_256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C) {
+define <8 x i32> @test_int_x86_avx2_vpdpwsud_256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C) {
 ; CHECK-LABEL: test_int_x86_avx2_vpdpwsud_256:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpdpwsud %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x76,0xd2,0xc2]
@@ -29,12 +29,12 @@ define <8 x i32> @test_int_x86_avx2_vpdpwsud_256(<8 x i32> %A, <8 x i32> %B, <8
 ; AVX10:       # %bb.0:
 ; AVX10-NEXT:    vpdpwsud %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x76,0x28,0xd2,0xc2]
 ; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
   ret <8 x i32> %ret
 }
-declare <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+declare <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
 
-define <4 x i32> @test_int_x86_avx2_vpdpwsuds_128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C) {
+define <4 x i32> @test_int_x86_avx2_vpdpwsuds_128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C) {
 ; CHECK-LABEL: test_int_x86_avx2_vpdpwsuds_128:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpdpwsuds %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x72,0xd3,0xc2]
@@ -44,12 +44,12 @@ define <4 x i32> @test_int_x86_avx2_vpdpwsuds_128(<4 x i32> %A, <4 x i32> %B, <4
 ; AVX10:       # %bb.0:
 ; AVX10-NEXT:    vpdpwsuds %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x76,0x08,0xd3,0xc2]
 ; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
   ret <4 x i32> %ret
 }
-declare <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+declare <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
 
-define <8 x i32> @test_int_x86_avx2_vpdpwsuds_256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C) {
+define <8 x i32> @test_int_x86_avx2_vpdpwsuds_256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C) {
 ; CHECK-LABEL: test_int_x86_avx2_vpdpwsuds_256:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpdpwsuds %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x76,0xd3,0xc2]
@@ -59,12 +59,12 @@ define <8 x i32> @test_int_x86_avx2_vpdpwsuds_256(<8 x i32> %A, <8 x i32> %B, <8
 ; AVX10:       # %bb.0:
 ; AVX10-NEXT:    vpdpwsuds %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x76,0x28,0xd3,0xc2]
 ; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
   ret <8 x i32> %ret
 }
-declare <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+declare <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
 
-define <4 x i32> @test_int_x86_avx2_vpdpwusd_128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C) {
+define <4 x i32> @test_int_x86_avx2_vpdpwusd_128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C) {
 ; CHECK-LABEL: test_int_x86_avx2_vpdpwusd_128:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpdpwusd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xd2,0xc2]
@@ -74,12 +74,12 @@ define <4 x i32> @test_int_x86_avx2_vpdpwusd_128(<4 x i32> %A, <4 x i32> %B, <4
 ; AVX10:       # %bb.0:
 ; AVX10-NEXT:    vpdpwusd %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x75,0x08,0xd2,0xc2]
 ; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
   ret <4 x i32> %ret
 }
-declare <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+declare <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
 
-define <8 x i32> @test_int_x86_avx2_vpdpwusd_256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C) {
+define <8 x i32> @test_int_x86_avx2_vpdpwusd_256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C) {
 ; CHECK-LABEL: test_int_x86_avx2_vpdpwusd_256:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpdpwusd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0xd2,0xc2]
@@ -89,12 +89,12 @@ define <8 x i32> @test_int_x86_avx2_vpdpwusd_256(<8 x i32> %A, <8 x i32> %B, <8
 ; AVX10:       # %bb.0:
 ; AVX10-NEXT:    vpdpwusd %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x75,0x28,0xd2,0xc2]
 ; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
   ret <8 x i32> %ret
 }
-declare <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+declare <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
 
-define <4 x i32> @test_int_x86_avx2_vpdpwusds_128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C) {
+define <4 x i32> @test_int_x86_avx2_vpdpwusds_128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C) {
 ; CHECK-LABEL: test_int_x86_avx2_vpdpwusds_128:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpdpwusds %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xd3,0xc2]
@@ -104,12 +104,12 @@ define <4 x i32> @test_int_x86_avx2_vpdpwusds_128(<4 x i32> %A, <4 x i32> %B, <4
 ; AVX10:       # %bb.0:
 ; AVX10-NEXT:    vpdpwusds %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x75,0x08,0xd3,0xc2]
 ; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
   ret <4 x i32> %ret
 }
-declare <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+declare <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
 
-define <8 x i32> @test_int_x86_avx2_vpdpwusds_256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C) {
+define <8 x i32> @test_int_x86_avx2_vpdpwusds_256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C) {
 ; CHECK-LABEL: test_int_x86_avx2_vpdpwusds_256:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpdpwusds %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0xd3,0xc2]
@@ -119,12 +119,12 @@ define <8 x i32> @test_int_x86_avx2_vpdpwusds_256(<8 x i32> %A, <8 x i32> %B, <8
 ; AVX10:       # %bb.0:
 ; AVX10-NEXT:    vpdpwusds %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x75,0x28,0xd3,0xc2]
 ; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
   ret <8 x i32> %ret
 }
-declare <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+declare <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
 
-define <4 x i32> @test_int_x86_avx2_vpdpwuud_128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C) {
+define <4 x i32> @test_int_x86_avx2_vpdpwuud_128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C) {
 ; CHECK-LABEL: test_int_x86_avx2_vpdpwuud_128:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpdpwuud %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x70,0xd2,0xc2]
@@ -134,12 +134,12 @@ define <4 x i32> @test_int_x86_avx2_vpdpwuud_128(<4 x i32> %A, <4 x i32> %B, <4
 ; AVX10:       # %bb.0:
 ; AVX10-NEXT:    vpdpwuud %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x74,0x08,0xd2,0xc2]
 ; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
   ret <4 x i32> %ret
 }
-declare <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+declare <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
 
-define <8 x i32> @test_int_x86_avx2_vpdpwuud_256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C) {
+define <8 x i32> @test_int_x86_avx2_vpdpwuud_256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C) {
 ; CHECK-LABEL: test_int_x86_avx2_vpdpwuud_256:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpdpwuud %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x74,0xd2,0xc2]
@@ -149,12 +149,12 @@ define <8 x i32> @test_int_x86_avx2_vpdpwuud_256(<8 x i32> %A, <8 x i32> %B, <8
 ; AVX10:       # %bb.0:
 ; AVX10-NEXT:    vpdpwuud %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x74,0x28,0xd2,0xc2]
 ; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
   ret <8 x i32> %ret
 }
-declare <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+declare <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
 
-define <4 x i32> @test_int_x86_avx2_vpdpwuuds_128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C) {
+define <4 x i32> @test_int_x86_avx2_vpdpwuuds_128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C) {
 ; CHECK-LABEL: test_int_x86_avx2_vpdpwuuds_128:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpdpwuuds %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x70,0xd3,0xc2]
@@ -164,12 +164,12 @@ define <4 x i32> @test_int_x86_avx2_vpdpwuuds_128(<4 x i32> %A, <4 x i32> %B, <4
 ; AVX10:       # %bb.0:
 ; AVX10-NEXT:    vpdpwuuds %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x74,0x08,0xd3,0xc2]
 ; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
   ret <4 x i32> %ret
 }
-declare <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+declare <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
 
-define <8 x i32> @test_int_x86_avx2_vpdpwuuds_256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C) {
+define <8 x i32> @test_int_x86_avx2_vpdpwuuds_256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C) {
 ; CHECK-LABEL: test_int_x86_avx2_vpdpwuuds_256:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpdpwuuds %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x74,0xd3,0xc2]
@@ -179,7 +179,7 @@ define <8 x i32> @test_int_x86_avx2_vpdpwuuds_256(<8 x i32> %A, <8 x i32> %B, <8
 ; AVX10:       # %bb.0:
 ; AVX10-NEXT:    vpdpwuuds %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x74,0x28,0xd3,0xc2]
 ; AVX10-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
   ret <8 x i32> %ret
 }
-declare <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+declare <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
diff --git a/llvm/test/CodeGen/X86/stack-folding-int-avxvnniint16.ll b/llvm/test/CodeGen/X86/stack-folding-int-avxvnniint16.ll
index 534352f322001..47537c8c4fb6e 100644
--- a/llvm/test/CodeGen/X86/stack-folding-int-avxvnniint16.ll
+++ b/llvm/test/CodeGen/X86/stack-folding-int-avxvnniint16.ll
@@ -1,10 +1,10 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -O3 -disable-peephole -verify-machineinstrs -mtriple=x86_64-unknown-unknown --show-mc-encoding -mattr=+avxvnniint16 | FileCheck %s
 
-declare <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
-declare <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
-declare <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
-declare <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+declare <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
+declare <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
+declare <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
+declare <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
 declare <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
 declare <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
 declare <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
@@ -14,7 +14,7 @@ declare <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> %A, <8 x i32> %B, <8 x i
 declare <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
 declare <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
 
-define <4 x i32> @test_int_x86_avx2_vpdpwsud_128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C) {
+define <4 x i32> @test_int_x86_avx2_vpdpwsud_128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C) {
 ; CHECK-LABEL: test_int_x86_avx2_vpdpwsud_128:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
@@ -26,11 +26,11 @@ define <4 x i32> @test_int_x86_avx2_vpdpwsud_128(<4 x i32> %A, <4 x i32> %B, <4
 ; CHECK-NEXT:    # encoding: [0xc4,0xe2,0x72,0xd2,0x44,0x24,0xe8]
 ; CHECK-NEXT:    retq # encoding: [0xc3]
   %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
-  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
   ret <4 x i32> %ret
 }
 
-define <8 x i32> @test_int_x86_avx2_vpdpwsud_256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C) {
+define <8 x i32> @test_int_x86_avx2_vpdpwsud_256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C) {
 ; CHECK-LABEL: test_int_x86_avx2_vpdpwsud_256:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -42,11 +42,11 @@ define <8 x i32> @test_int_x86_avx2_vpdpwsud_256(<8 x i32> %A, <8 x i32> %B, <8
 ; CHECK-NEXT:    # encoding: [0xc4,0xe2,0x76,0xd2,0x44,0x24,0xd8]
 ; CHECK-NEXT:    retq # encoding: [0xc3]
   %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
-  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
   ret <8 x i32> %ret
 }
 
-define <4 x i32> @test_int_x86_avx2_vpdpwsuds_128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C) {
+define <4 x i32> @test_int_x86_avx2_vpdpwsuds_128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C) {
 ; CHECK-LABEL: test_int_x86_avx2_vpdpwsuds_128:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
@@ -58,11 +58,11 @@ define <4 x i32> @test_int_x86_avx2_vpdpwsuds_128(<4 x i32> %A, <4 x i32> %B, <4
 ; CHECK-NEXT:    # encoding: [0xc4,0xe2,0x72,0xd3,0x44,0x24,0xe8]
 ; CHECK-NEXT:    retq # encoding: [0xc3]
   %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
-  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
   ret <4 x i32> %ret
 }
 
-define <8 x i32> @test_int_x86_avx2_vpdpwsuds_256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C) {
+define <8 x i32> @test_int_x86_avx2_vpdpwsuds_256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C) {
 ; CHECK-LABEL: test_int_x86_avx2_vpdpwsuds_256:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -74,7 +74,7 @@ define <8 x i32> @test_int_x86_avx2_vpdpwsuds_256(<8 x i32> %A, <8 x i32> %B, <8
 ; CHECK-NEXT:    # encoding: [0xc4,0xe2,0x76,0xd3,0x44,0x24,0xd8]
 ; CHECK-NEXT:    retq # encoding: [0xc3]
   %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{flags}"()
-  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
   ret <8 x i32> %ret
 }
 
diff --git a/llvm/test/Instrumentation/MemorySanitizer/X86/avx10_2_512ni-intrinsics.ll b/llvm/test/Instrumentation/MemorySanitizer/X86/avx10_2_512ni-intrinsics.ll
index 8900085af030d..9d5cabca5fef8 100644
--- a/llvm/test/Instrumentation/MemorySanitizer/X86/avx10_2_512ni-intrinsics.ll
+++ b/llvm/test/Instrumentation/MemorySanitizer/X86/avx10_2_512ni-intrinsics.ll
@@ -497,12 +497,12 @@ declare <16 x i32> @llvm.x86.avx10.vpdpbuud.512(<16 x i32>, <64 x i8>, <64 x i8>
 declare <16 x i32> @llvm.x86.avx10.vpdpbuuds.512(<16 x i32>, <64 x i8>, <64 x i8>)
 
 
-define <16 x i32> @test_mm512_dpwsud_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr %pB) sanitize_memory {
+define <16 x i32> @test_mm512_dpwsud_epi32(<16 x i32> %__W, <32 x i16> %__A, ptr %pB) sanitize_memory {
 ; CHECK-LABEL: define <16 x i32> @test_mm512_dpwsud_epi32(
-; CHECK-SAME: <16 x i32> [[__W:%.*]], <16 x i32> [[__A:%.*]], ptr [[PB:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: <16 x i32> [[__W:%.*]], <32 x i16> [[__A:%.*]], ptr [[PB:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr getelementptr (i8, ptr @__msan_param_tls, i64 128), align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
 ; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
 ; CHECK-NEXT:    br i1 [[_MSCMP]], label %[[BB4:.*]], label %[[BB5:.*]], !prof [[PROF1]]
@@ -510,87 +510,123 @@ define <16 x i32> @test_mm512_dpwsud_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr
 ; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR5]]
 ; CHECK-NEXT:    unreachable
 ; CHECK:       [[BB5]]:
-; CHECK-NEXT:    [[__B:%.*]] = load <16 x i32>, ptr [[PB]], align 64
+; CHECK-NEXT:    [[__B:%.*]] = load <32 x i16>, ptr [[PB]], align 64
 ; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PB]] to i64
 ; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
 ; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
-; CHECK-NEXT:    [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP8]], align 64
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <16 x i32> [[TMP2]], [[TMP3]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[_MSPROP]], [[_MSLD]]
-; CHECK-NEXT:    [[RES:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32> [[__W]], <16 x i32> [[__A]], <16 x i32> [[__B]])
-; CHECK-NEXT:    store <16 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP8]], align 64
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <32 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne <32 x i16> [[_MSLD]], zeroinitializer
+; CHECK-NEXT:    [[TMP11:%.*]] = icmp ne <32 x i16> [[__A]], zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = icmp ne <32 x i16> [[__B]], zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = and <32 x i1> [[TMP9]], [[TMP10]]
+; CHECK-NEXT:    [[TMP14:%.*]] = and <32 x i1> [[TMP11]], [[TMP10]]
+; CHECK-NEXT:    [[TMP15:%.*]] = and <32 x i1> [[TMP9]], [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <32 x i1> [[TMP13]], [[TMP14]]
+; CHECK-NEXT:    [[TMP17:%.*]] = or <32 x i1> [[TMP16]], [[TMP15]]
+; CHECK-NEXT:    [[TMP18:%.*]] = sext <32 x i1> [[TMP17]] to <32 x i16>
+; CHECK-NEXT:    [[TMP19:%.*]] = bitcast <32 x i16> [[TMP18]] to <16 x i32>
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ne <16 x i32> [[TMP19]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = sext <16 x i1> [[TMP20]] to <16 x i32>
+; CHECK-NEXT:    [[TMP22:%.*]] = or <16 x i32> [[TMP21]], [[TMP4]]
+; CHECK-NEXT:    [[RES:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32> [[__W]], <32 x i16> [[__A]], <32 x i16> [[__B]])
+; CHECK-NEXT:    store <16 x i32> [[TMP22]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <16 x i32> [[RES]]
 ;
-  %__B = load <16 x i32>, ptr %pB
-  %res = tail call <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B)
+  %__B = load <32 x i16>, ptr %pB
+  %res = tail call <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32> %__W, <32 x i16> %__A, <32 x i16> %__B)
   ret <16 x i32> %res
 }
 
-define <16 x i32> @test_mm512_mask_dpwsuds_epi32(<16 x i32> %__W, i16 zeroext %__U, <16 x i32> %__A, <16 x i32> %__B) sanitize_memory {
+define <16 x i32> @test_mm512_mask_dpwsuds_epi32(<16 x i32> %__W, i16 zeroext %__U, <32 x i16> %__A, <32 x i16> %__B) sanitize_memory {
 ; CHECK-LABEL: define <16 x i32> @test_mm512_mask_dpwsuds_epi32(
-; CHECK-SAME: <16 x i32> [[__W:%.*]], i16 zeroext [[__U:%.*]], <16 x i32> [[__A:%.*]], <16 x i32> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: <16 x i32> [[__W:%.*]], i16 zeroext [[__U:%.*]], <32 x i16> [[__A:%.*]], <32 x i16> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 136), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 136), align 8
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i16, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <16 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpwsuds.512(<16 x i32> [[__W]], <16 x i32> [[__A]], <16 x i32> [[__B]])
+; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <32 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ne <32 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ne <32 x i16> [[__A]], zeroinitializer
+; CHECK-NEXT:    [[TMP22:%.*]] = icmp ne <32 x i16> [[__B]], zeroinitializer
+; CHECK-NEXT:    [[TMP23:%.*]] = and <32 x i1> [[TMP19]], [[TMP20]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <32 x i1> [[TMP21]], [[TMP20]]
+; CHECK-NEXT:    [[TMP11:%.*]] = and <32 x i1> [[TMP19]], [[TMP22]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <32 x i1> [[TMP23]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = or <32 x i1> [[TMP12]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = sext <32 x i1> [[TMP13]] to <32 x i16>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast <32 x i16> [[TMP14]] to <16 x i32>
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <16 x i32> [[TMP15]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = sext <16 x i1> [[TMP16]] to <16 x i32>
+; CHECK-NEXT:    [[TMP18:%.*]] = or <16 x i32> [[TMP17]], [[TMP1]]
+; CHECK-NEXT:    [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpwsuds.512(<16 x i32> [[__W]], <32 x i16> [[__A]], <32 x i16> [[__B]])
 ; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i16 [[TMP4]] to <16 x i1>
 ; CHECK-NEXT:    [[BST:%.*]] = bitcast i16 [[__U]] to <16 x i1>
-; CHECK-NEXT:    [[TMP6:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[_MSPROP1]], <16 x i32> [[TMP1]]
+; CHECK-NEXT:    [[TMP6:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[TMP18]], <16 x i32> [[TMP1]]
 ; CHECK-NEXT:    [[TMP7:%.*]] = xor <16 x i32> [[DPI]], [[__W]]
-; CHECK-NEXT:    [[TMP8:%.*]] = or <16 x i32> [[TMP7]], [[_MSPROP1]]
+; CHECK-NEXT:    [[TMP8:%.*]] = or <16 x i32> [[TMP7]], [[TMP18]]
 ; CHECK-NEXT:    [[TMP9:%.*]] = or <16 x i32> [[TMP8]], [[TMP1]]
 ; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <16 x i1> [[TMP5]], <16 x i32> [[TMP9]], <16 x i32> [[TMP6]]
 ; CHECK-NEXT:    [[RES:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[DPI]], <16 x i32> [[__W]]
 ; CHECK-NEXT:    store <16 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <16 x i32> [[RES]]
 ;
-  %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpwsuds.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B)
+  %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpwsuds.512(<16 x i32> %__W, <32 x i16> %__A, <32 x i16> %__B)
   %bst = bitcast i16 %__U to <16 x i1>
   %res = select <16 x i1> %bst, <16 x i32> %dpi, <16 x i32> %__W
   ret <16 x i32> %res
 }
 
-define <16 x i32> @test_mm512_maskz_dpwsud_epi32(i16 zeroext %__U, <16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) sanitize_memory {
+define <16 x i32> @test_mm512_maskz_dpwsud_epi32(i16 zeroext %__U, <16 x i32> %__W, <32 x i16> %__A, <32 x i16> %__B) sanitize_memory {
 ; CHECK-LABEL: define <16 x i32> @test_mm512_maskz_dpwsud_epi32(
-; CHECK-SAME: i16 zeroext [[__U:%.*]], <16 x i32> [[__W:%.*]], <16 x i32> [[__A:%.*]], <16 x i32> [[__B:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 8), align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 136), align 8
+; CHECK-SAME: i16 zeroext [[__U:%.*]], <16 x i32> [[__W:%.*]], <32 x i16> [[__A:%.*]], <32 x i16> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 136), align 8
+; CHECK-NEXT:    [[TMP19:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 8), align 8
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i16, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <16 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32> [[__W]], <16 x i32> [[__A]], <16 x i32> [[__B]])
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ne <32 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ne <32 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP22:%.*]] = icmp ne <32 x i16> [[__A]], zeroinitializer
+; CHECK-NEXT:    [[TMP23:%.*]] = icmp ne <32 x i16> [[__B]], zeroinitializer
+; CHECK-NEXT:    [[TMP24:%.*]] = and <32 x i1> [[TMP20]], [[TMP21]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <32 x i1> [[TMP22]], [[TMP21]]
+; CHECK-NEXT:    [[TMP11:%.*]] = and <32 x i1> [[TMP20]], [[TMP23]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <32 x i1> [[TMP24]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = or <32 x i1> [[TMP12]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = sext <32 x i1> [[TMP13]] to <32 x i16>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast <32 x i16> [[TMP14]] to <16 x i32>
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <16 x i32> [[TMP15]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = sext <16 x i1> [[TMP16]] to <16 x i32>
+; CHECK-NEXT:    [[TMP18:%.*]] = or <16 x i32> [[TMP17]], [[TMP19]]
+; CHECK-NEXT:    [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32> [[__W]], <32 x i16> [[__A]], <32 x i16> [[__B]])
 ; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i16 [[TMP4]] to <16 x i1>
 ; CHECK-NEXT:    [[BST:%.*]] = bitcast i16 [[__U]] to <16 x i1>
-; CHECK-NEXT:    [[TMP6:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[_MSPROP1]], <16 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[TMP18]], <16 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP7:%.*]] = xor <16 x i32> [[DPI]], zeroinitializer
-; CHECK-NEXT:    [[TMP8:%.*]] = or <16 x i32> [[TMP7]], [[_MSPROP1]]
+; CHECK-NEXT:    [[TMP8:%.*]] = or <16 x i32> [[TMP7]], [[TMP18]]
 ; CHECK-NEXT:    [[TMP9:%.*]] = or <16 x i32> [[TMP8]], zeroinitializer
 ; CHECK-NEXT:    [[_MSPROP_SELECT:%.*]] = select <16 x i1> [[TMP5]], <16 x i32> [[TMP9]], <16 x i32> [[TMP6]]
 ; CHECK-NEXT:    [[RES:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[DPI]], <16 x i32> zeroinitializer
 ; CHECK-NEXT:    store <16 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <16 x i32> [[RES]]
 ;
-  %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B)
+  %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32> %__W, <32 x i16> %__A, <32 x i16> %__B)
   %bst = bitcast i16 %__U to <16 x i1>
   %res = select <16 x i1> %bst, <16 x i32> %dpi, <16 x i32> zeroinitializer
   ret <16 x i32> %res
 }
 
-declare <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32>, <16 x i32>, <16 x i32>)
-declare <16 x i32> @llvm.x86.avx10.vpdpwsuds.512(<16 x i32>, <16 x i32>, <16 x i32>)
+declare <16 x i32> @llvm.x86.avx10.vpdpwsud.512(<16 x i32>, <32 x i16>, <32 x i16>)
+declare <16 x i32> @llvm.x86.avx10.vpdpwsuds.512(<16 x i32>, <32 x i16>, <32 x i16>)
 
-define <16 x i32> @test_mm512_dpwusd_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr %pB) sanitize_memory {
+define <16 x i32> @test_mm512_dpwusd_epi32(<16 x i32> %__W, <32 x i16> %__A, ptr %pB) sanitize_memory {
 ; CHECK-LABEL: define <16 x i32> @test_mm512_dpwusd_epi32(
-; CHECK-SAME: <16 x i32> [[__W:%.*]], <16 x i32> [[__A:%.*]], ptr [[PB:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: <16 x i32> [[__W:%.*]], <32 x i16> [[__A:%.*]], ptr [[PB:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr getelementptr (i8, ptr @__msan_param_tls, i64 128), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
 ; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
 ; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
 ; CHECK-NEXT:    br i1 [[_MSCMP]], label %[[BB4:.*]], label %[[BB5:.*]], !prof [[PROF1]]
@@ -598,33 +634,57 @@ define <16 x i32> @test_mm512_dpwusd_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr
 ; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR5]]
 ; CHECK-NEXT:    unreachable
 ; CHECK:       [[BB5]]:
-; CHECK-NEXT:    [[__B:%.*]] = load <16 x i32>, ptr [[PB]], align 64
+; CHECK-NEXT:    [[__B:%.*]] = load <32 x i16>, ptr [[PB]], align 64
 ; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PB]] to i64
 ; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
 ; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
-; CHECK-NEXT:    [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP8]], align 64
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <16 x i32> [[TMP2]], [[TMP3]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[_MSPROP]], [[_MSLD]]
-; CHECK-NEXT:    [[RES:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpwusd.512(<16 x i32> [[__W]], <16 x i32> [[__A]], <16 x i32> [[__B]])
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP8]], align 64
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <32 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne <32 x i16> [[_MSLD]], zeroinitializer
+; CHECK-NEXT:    [[TMP11:%.*]] = icmp ne <32 x i16> [[__A]], zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = icmp ne <32 x i16> [[__B]], zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = and <32 x i1> [[TMP9]], [[TMP10]]
+; CHECK-NEXT:    [[TMP14:%.*]] = and <32 x i1> [[TMP11]], [[TMP10]]
+; CHECK-NEXT:    [[TMP15:%.*]] = and <32 x i1> [[TMP9]], [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <32 x i1> [[TMP13]], [[TMP14]]
+; CHECK-NEXT:    [[TMP17:%.*]] = or <32 x i1> [[TMP16]], [[TMP15]]
+; CHECK-NEXT:    [[TMP18:%.*]] = sext <32 x i1> [[TMP17]] to <32 x i16>
+; CHECK-NEXT:    [[TMP19:%.*]] = bitcast <32 x i16> [[TMP18]] to <16 x i32>
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ne <16 x i32> [[TMP19]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = sext <16 x i1> [[TMP20]] to <16 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[TMP21]], [[TMP2]]
+; CHECK-NEXT:    [[RES:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpwusd.512(<16 x i32> [[__W]], <32 x i16> [[__A]], <32 x i16> [[__B]])
 ; CHECK-NEXT:    store <16 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <16 x i32> [[RES]]
 ;
-  %__B = load <16 x i32>, ptr %pB
-  %res = tail call <16 x i32> @llvm.x86.avx10.vpdpwusd.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B)
+  %__B = load <32 x i16>, ptr %pB
+  %res = tail call <16 x i32> @llvm.x86.avx10.vpdpwusd.512(<16 x i32> %__W, <32 x i16> %__A, <32 x i16> %__B)
   ret <16 x i32> %res
 }
 
-define <16 x i32> @test_mm512_mask_dpwusds_epi32(<16 x i32> %__W, i16 zeroext %__U, <16 x i32> %__A, <16 x i32> %__B) sanitize_memory {
+define <16 x i32> @test_mm512_mask_dpwusds_epi32(<16 x i32> %__W, i16 zeroext %__U, <32 x i16> %__A, <32 x i16> %__B) sanitize_memory {
 ; CHECK-LABEL: define <16 x i32> @test_mm512_mask_dpwusds_epi32(
-; CHECK-SAME: <16 x i32> [[__W:%.*]], i16 zeroext [[__U:%.*]], <16 x i32> [[__A:%.*]], <16 x i32> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: <16 x i32> [[__W:%.*]], i16 zeroext [[__U:%.*]], <32 x i16> [[__A:%.*]], <32 x i16> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 136), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 136), align 8
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i16, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <16 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpwusds.512(<16 x i32> [[__W]], <16 x i32> [[__A]], <16 x i32> [[__B]])
+; CHECK-NEXT:    [[TMP18:%.*]] = icmp ne <32 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <32 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ne <32 x i16> [[__A]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ne <32 x i16> [[__B]], zeroinitializer
+; CHECK-NEXT:    [[TMP22:%.*]] = and <32 x i1> [[TMP18]], [[TMP19]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <32 x i1> [[TMP20]], [[TMP19]]
+; CHECK-NEXT:    [[TMP11:%.*]] = and <32 x i1> [[TMP18]], [[TMP21]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <32 x i1> [[TMP22]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = or <32 x i1> [[TMP12]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = sext <32 x i1> [[TMP13]] to <32 x i16>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast <32 x i16> [[TMP14]] to <16 x i32>
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <16 x i32> [[TMP15]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = sext <16 x i1> [[TMP16]] to <16 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[TMP17]], [[TMP1]]
+; CHECK-NEXT:    [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpwusds.512(<16 x i32> [[__W]], <32 x i16> [[__A]], <32 x i16> [[__B]])
 ; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i16 [[TMP4]] to <16 x i1>
 ; CHECK-NEXT:    [[BST:%.*]] = bitcast i16 [[__U]] to <16 x i1>
 ; CHECK-NEXT:    [[TMP6:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[_MSPROP1]], <16 x i32> [[TMP1]]
@@ -636,23 +696,35 @@ define <16 x i32> @test_mm512_mask_dpwusds_epi32(<16 x i32> %__W, i16 zeroext %_
 ; CHECK-NEXT:    store <16 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <16 x i32> [[RES]]
 ;
-  %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpwusds.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B)
+  %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpwusds.512(<16 x i32> %__W, <32 x i16> %__A, <32 x i16> %__B)
   %bst = bitcast i16 %__U to <16 x i1>
   %res = select <16 x i1> %bst, <16 x i32> %dpi, <16 x i32> %__W
   ret <16 x i32> %res
 }
 
-define <16 x i32> @test_mm512_maskz_dpwusd_epi32(i16 zeroext %__U, <16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) sanitize_memory {
+define <16 x i32> @test_mm512_maskz_dpwusd_epi32(i16 zeroext %__U, <16 x i32> %__W, <32 x i16> %__A, <32 x i16> %__B) sanitize_memory {
 ; CHECK-LABEL: define <16 x i32> @test_mm512_maskz_dpwusd_epi32(
-; CHECK-SAME: i16 zeroext [[__U:%.*]], <16 x i32> [[__W:%.*]], <16 x i32> [[__A:%.*]], <16 x i32> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: i16 zeroext [[__U:%.*]], <16 x i32> [[__W:%.*]], <32 x i16> [[__A:%.*]], <32 x i16> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 136), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 8), align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 136), align 8
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i16, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <16 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpwusd.512(<16 x i32> [[__W]], <16 x i32> [[__A]], <16 x i32> [[__B]])
+; CHECK-NEXT:    [[TMP18:%.*]] = icmp ne <32 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <32 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ne <32 x i16> [[__A]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ne <32 x i16> [[__B]], zeroinitializer
+; CHECK-NEXT:    [[TMP22:%.*]] = and <32 x i1> [[TMP18]], [[TMP19]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <32 x i1> [[TMP20]], [[TMP19]]
+; CHECK-NEXT:    [[TMP11:%.*]] = and <32 x i1> [[TMP18]], [[TMP21]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <32 x i1> [[TMP22]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = or <32 x i1> [[TMP12]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = sext <32 x i1> [[TMP13]] to <32 x i16>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast <32 x i16> [[TMP14]] to <16 x i32>
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <16 x i32> [[TMP15]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = sext <16 x i1> [[TMP16]] to <16 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[TMP17]], [[TMP1]]
+; CHECK-NEXT:    [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpwusd.512(<16 x i32> [[__W]], <32 x i16> [[__A]], <32 x i16> [[__B]])
 ; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i16 [[TMP4]] to <16 x i1>
 ; CHECK-NEXT:    [[BST:%.*]] = bitcast i16 [[__U]] to <16 x i1>
 ; CHECK-NEXT:    [[TMP6:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[_MSPROP1]], <16 x i32> zeroinitializer
@@ -664,21 +736,21 @@ define <16 x i32> @test_mm512_maskz_dpwusd_epi32(i16 zeroext %__U, <16 x i32> %_
 ; CHECK-NEXT:    store <16 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <16 x i32> [[RES]]
 ;
-  %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpwusd.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B)
+  %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpwusd.512(<16 x i32> %__W, <32 x i16> %__A, <32 x i16> %__B)
   %bst = bitcast i16 %__U to <16 x i1>
   %res = select <16 x i1> %bst, <16 x i32> %dpi, <16 x i32> zeroinitializer
   ret <16 x i32> %res
 }
 
-declare <16 x i32> @llvm.x86.avx10.vpdpwusd.512(<16 x i32>, <16 x i32>, <16 x i32>)
-declare <16 x i32> @llvm.x86.avx10.vpdpwusds.512(<16 x i32>, <16 x i32>, <16 x i32>)
+declare <16 x i32> @llvm.x86.avx10.vpdpwusd.512(<16 x i32>, <32 x i16>, <32 x i16>)
+declare <16 x i32> @llvm.x86.avx10.vpdpwusds.512(<16 x i32>, <32 x i16>, <32 x i16>)
 
-define <16 x i32> @test_mm512_dpwuud_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr %pB) sanitize_memory {
+define <16 x i32> @test_mm512_dpwuud_epi32(<16 x i32> %__W, <32 x i16> %__A, ptr %pB) sanitize_memory {
 ; CHECK-LABEL: define <16 x i32> @test_mm512_dpwuud_epi32(
-; CHECK-SAME: <16 x i32> [[__W:%.*]], <16 x i32> [[__A:%.*]], ptr [[PB:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: <16 x i32> [[__W:%.*]], <32 x i16> [[__A:%.*]], ptr [[PB:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr getelementptr (i8, ptr @__msan_param_tls, i64 128), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
 ; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
 ; CHECK-NEXT:    [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0
 ; CHECK-NEXT:    br i1 [[_MSCMP]], label %[[BB4:.*]], label %[[BB5:.*]], !prof [[PROF1]]
@@ -686,33 +758,57 @@ define <16 x i32> @test_mm512_dpwuud_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr
 ; CHECK-NEXT:    call void @__msan_warning_noreturn() #[[ATTR5]]
 ; CHECK-NEXT:    unreachable
 ; CHECK:       [[BB5]]:
-; CHECK-NEXT:    [[__B:%.*]] = load <16 x i32>, ptr [[PB]], align 64
+; CHECK-NEXT:    [[__B:%.*]] = load <32 x i16>, ptr [[PB]], align 64
 ; CHECK-NEXT:    [[TMP6:%.*]] = ptrtoint ptr [[PB]] to i64
 ; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080
 ; CHECK-NEXT:    [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr
-; CHECK-NEXT:    [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP8]], align 64
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <16 x i32> [[TMP2]], [[TMP3]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[_MSPROP]], [[_MSLD]]
-; CHECK-NEXT:    [[RES:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpwuud.512(<16 x i32> [[__W]], <16 x i32> [[__A]], <16 x i32> [[__B]])
+; CHECK-NEXT:    [[_MSLD:%.*]] = load <32 x i16>, ptr [[TMP8]], align 64
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp ne <32 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP10:%.*]] = icmp ne <32 x i16> [[_MSLD]], zeroinitializer
+; CHECK-NEXT:    [[TMP11:%.*]] = icmp ne <32 x i16> [[__A]], zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = icmp ne <32 x i16> [[__B]], zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = and <32 x i1> [[TMP9]], [[TMP10]]
+; CHECK-NEXT:    [[TMP14:%.*]] = and <32 x i1> [[TMP11]], [[TMP10]]
+; CHECK-NEXT:    [[TMP15:%.*]] = and <32 x i1> [[TMP9]], [[TMP12]]
+; CHECK-NEXT:    [[TMP16:%.*]] = or <32 x i1> [[TMP13]], [[TMP14]]
+; CHECK-NEXT:    [[TMP17:%.*]] = or <32 x i1> [[TMP16]], [[TMP15]]
+; CHECK-NEXT:    [[TMP18:%.*]] = sext <32 x i1> [[TMP17]] to <32 x i16>
+; CHECK-NEXT:    [[TMP19:%.*]] = bitcast <32 x i16> [[TMP18]] to <16 x i32>
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ne <16 x i32> [[TMP19]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = sext <16 x i1> [[TMP20]] to <16 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[TMP21]], [[TMP2]]
+; CHECK-NEXT:    [[RES:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpwuud.512(<16 x i32> [[__W]], <32 x i16> [[__A]], <32 x i16> [[__B]])
 ; CHECK-NEXT:    store <16 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <16 x i32> [[RES]]
 ;
-  %__B = load <16 x i32>, ptr %pB
-  %res = tail call <16 x i32> @llvm.x86.avx10.vpdpwuud.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B)
+  %__B = load <32 x i16>, ptr %pB
+  %res = tail call <16 x i32> @llvm.x86.avx10.vpdpwuud.512(<16 x i32> %__W, <32 x i16> %__A, <32 x i16> %__B)
   ret <16 x i32> %res
 }
 
-define <16 x i32> @test_mm512_mask_dpwuuds_epi32(<16 x i32> %__W, i16 zeroext %__U, <16 x i32> %__A, <16 x i32> %__B) sanitize_memory {
+define <16 x i32> @test_mm512_mask_dpwuuds_epi32(<16 x i32> %__W, i16 zeroext %__U, <32 x i16> %__A, <32 x i16> %__B) sanitize_memory {
 ; CHECK-LABEL: define <16 x i32> @test_mm512_mask_dpwuuds_epi32(
-; CHECK-SAME: <16 x i32> [[__W:%.*]], i16 zeroext [[__U:%.*]], <16 x i32> [[__A:%.*]], <16 x i32> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: <16 x i32> [[__W:%.*]], i16 zeroext [[__U:%.*]], <32 x i16> [[__A:%.*]], <32 x i16> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 136), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 136), align 8
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i16, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <16 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpwuuds.512(<16 x i32> [[__W]], <16 x i32> [[__A]], <16 x i32> [[__B]])
+; CHECK-NEXT:    [[TMP18:%.*]] = icmp ne <32 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <32 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ne <32 x i16> [[__A]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ne <32 x i16> [[__B]], zeroinitializer
+; CHECK-NEXT:    [[TMP22:%.*]] = and <32 x i1> [[TMP18]], [[TMP19]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <32 x i1> [[TMP20]], [[TMP19]]
+; CHECK-NEXT:    [[TMP11:%.*]] = and <32 x i1> [[TMP18]], [[TMP21]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <32 x i1> [[TMP22]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = or <32 x i1> [[TMP12]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = sext <32 x i1> [[TMP13]] to <32 x i16>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast <32 x i16> [[TMP14]] to <16 x i32>
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <16 x i32> [[TMP15]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = sext <16 x i1> [[TMP16]] to <16 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[TMP17]], [[TMP1]]
+; CHECK-NEXT:    [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpwuuds.512(<16 x i32> [[__W]], <32 x i16> [[__A]], <32 x i16> [[__B]])
 ; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i16 [[TMP4]] to <16 x i1>
 ; CHECK-NEXT:    [[BST:%.*]] = bitcast i16 [[__U]] to <16 x i1>
 ; CHECK-NEXT:    [[TMP6:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[_MSPROP1]], <16 x i32> [[TMP1]]
@@ -724,23 +820,35 @@ define <16 x i32> @test_mm512_mask_dpwuuds_epi32(<16 x i32> %__W, i16 zeroext %_
 ; CHECK-NEXT:    store <16 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <16 x i32> [[RES]]
 ;
-  %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpwuuds.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B)
+  %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpwuuds.512(<16 x i32> %__W, <32 x i16> %__A, <32 x i16> %__B)
   %bst = bitcast i16 %__U to <16 x i1>
   %res = select <16 x i1> %bst, <16 x i32> %dpi, <16 x i32> %__W
   ret <16 x i32> %res
 }
 
-define <16 x i32> @test_mm512_maskz_dpwuud_epi32(i16 zeroext %__U, <16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) sanitize_memory {
+define <16 x i32> @test_mm512_maskz_dpwuud_epi32(i16 zeroext %__U, <16 x i32> %__W, <32 x i16> %__A, <32 x i16> %__B) sanitize_memory {
 ; CHECK-LABEL: define <16 x i32> @test_mm512_maskz_dpwuud_epi32(
-; CHECK-SAME: i16 zeroext [[__U:%.*]], <16 x i32> [[__W:%.*]], <16 x i32> [[__A:%.*]], <16 x i32> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: i16 zeroext [[__U:%.*]], <16 x i32> [[__W:%.*]], <32 x i16> [[__A:%.*]], <32 x i16> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <32 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 136), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 8), align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 136), align 8
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i16, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <16 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpwuud.512(<16 x i32> [[__W]], <16 x i32> [[__A]], <16 x i32> [[__B]])
+; CHECK-NEXT:    [[TMP18:%.*]] = icmp ne <32 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <32 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ne <32 x i16> [[__A]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ne <32 x i16> [[__B]], zeroinitializer
+; CHECK-NEXT:    [[TMP22:%.*]] = and <32 x i1> [[TMP18]], [[TMP19]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <32 x i1> [[TMP20]], [[TMP19]]
+; CHECK-NEXT:    [[TMP11:%.*]] = and <32 x i1> [[TMP18]], [[TMP21]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <32 x i1> [[TMP22]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = or <32 x i1> [[TMP12]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = sext <32 x i1> [[TMP13]] to <32 x i16>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast <32 x i16> [[TMP14]] to <16 x i32>
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <16 x i32> [[TMP15]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = sext <16 x i1> [[TMP16]] to <16 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <16 x i32> [[TMP17]], [[TMP1]]
+; CHECK-NEXT:    [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpwuud.512(<16 x i32> [[__W]], <32 x i16> [[__A]], <32 x i16> [[__B]])
 ; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i16 [[TMP4]] to <16 x i1>
 ; CHECK-NEXT:    [[BST:%.*]] = bitcast i16 [[__U]] to <16 x i1>
 ; CHECK-NEXT:    [[TMP6:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[_MSPROP1]], <16 x i32> zeroinitializer
@@ -752,14 +860,14 @@ define <16 x i32> @test_mm512_maskz_dpwuud_epi32(i16 zeroext %__U, <16 x i32> %_
 ; CHECK-NEXT:    store <16 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <16 x i32> [[RES]]
 ;
-  %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpwuud.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B)
+  %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpwuud.512(<16 x i32> %__W, <32 x i16> %__A, <32 x i16> %__B)
   %bst = bitcast i16 %__U to <16 x i1>
   %res = select <16 x i1> %bst, <16 x i32> %dpi, <16 x i32> zeroinitializer
   ret <16 x i32> %res
 }
 
-declare <16 x i32> @llvm.x86.avx10.vpdpwuud.512(<16 x i32>, <16 x i32>, <16 x i32>)
-declare <16 x i32> @llvm.x86.avx10.vpdpwuuds.512(<16 x i32>, <16 x i32>, <16 x i32>)
+declare <16 x i32> @llvm.x86.avx10.vpdpwuud.512(<16 x i32>, <32 x i16>, <32 x i16>)
+declare <16 x i32> @llvm.x86.avx10.vpdpwuuds.512(<16 x i32>, <32 x i16>, <32 x i16>)
 
 
 define { <32 x i16>, <32 x i16>, <32 x i16> } @test_mm512_mask_mpsadbw(<64 x i8> %x0, <64 x i8> %x1, <32 x i16> %x3, i32 %x4) sanitize_memory {
diff --git a/llvm/test/Instrumentation/MemorySanitizer/X86/avx10_2ni-intrinsics.ll b/llvm/test/Instrumentation/MemorySanitizer/X86/avx10_2ni-intrinsics.ll
index def7ba3f10770..6a53a1595271e 100644
--- a/llvm/test/Instrumentation/MemorySanitizer/X86/avx10_2ni-intrinsics.ll
+++ b/llvm/test/Instrumentation/MemorySanitizer/X86/avx10_2ni-intrinsics.ll
@@ -739,17 +739,29 @@ declare <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32>, <32 x i8>, <32 x i8>)
 declare <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32>, <32 x i8>, <32 x i8>)
 
 
-define <4 x i32> @test_mm_mask_dpwsud_epi32(<4 x i32> %__W, i4 zeroext %__U, <4 x i32> %__A, <4 x i32> %__B) sanitize_memory {
+define <4 x i32> @test_mm_mask_dpwsud_epi32(<4 x i32> %__W, i4 zeroext %__U, <8 x i16> %__A, <8 x i16> %__B) sanitize_memory {
 ; CHECK-LABEL: define <4 x i32> @test_mm_mask_dpwsud_epi32(
-; CHECK-SAME: <4 x i32> [[__W:%.*]], i4 zeroext [[__U:%.*]], <4 x i32> [[__A:%.*]], <4 x i32> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: <4 x i32> [[__W:%.*]], i4 zeroext [[__U:%.*]], <8 x i16> [[__A:%.*]], <8 x i16> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 24), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 24), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i4, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <4 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> [[__W]], <4 x i32> [[__A]], <4 x i32> [[__B]])
+; CHECK-NEXT:    [[TMP18:%.*]] = icmp ne <8 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <8 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ne <8 x i16> [[__A]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ne <8 x i16> [[__B]], zeroinitializer
+; CHECK-NEXT:    [[TMP22:%.*]] = and <8 x i1> [[TMP18]], [[TMP19]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <8 x i1> [[TMP20]], [[TMP19]]
+; CHECK-NEXT:    [[TMP11:%.*]] = and <8 x i1> [[TMP18]], [[TMP21]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <8 x i1> [[TMP22]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = or <8 x i1> [[TMP12]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = sext <8 x i1> [[TMP13]] to <8 x i16>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast <8 x i16> [[TMP14]] to <4 x i32>
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <4 x i32> [[TMP15]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = sext <4 x i1> [[TMP16]] to <4 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[TMP17]], [[TMP1]]
+; CHECK-NEXT:    [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> [[__W]], <8 x i16> [[__A]], <8 x i16> [[__B]])
 ; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i4 [[TMP4]] to <4 x i1>
 ; CHECK-NEXT:    [[BST:%.*]] = bitcast i4 [[__U]] to <4 x i1>
 ; CHECK-NEXT:    [[TMP6:%.*]] = select <4 x i1> [[BST]], <4 x i32> [[_MSPROP1]], <4 x i32> [[TMP1]]
@@ -761,23 +773,35 @@ define <4 x i32> @test_mm_mask_dpwsud_epi32(<4 x i32> %__W, i4 zeroext %__U, <4
 ; CHECK-NEXT:    store <4 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <4 x i32> [[RES]]
 ;
-  %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B)
+  %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %__W, <8 x i16> %__A, <8 x i16> %__B)
   %bst = bitcast i4 %__U to <4 x i1>
   %res = select <4 x i1> %bst, <4 x i32> %dpi, <4 x i32> %__W
   ret <4 x i32> %res
 }
 
-define <4 x i32> @test_mm_maskz_dpwsuds_epi32(i4 zeroext %__U, <4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B) sanitize_memory {
+define <4 x i32> @test_mm_maskz_dpwsuds_epi32(i4 zeroext %__U, <4 x i32> %__W, <8 x i16> %__A, <8 x i16> %__B) sanitize_memory {
 ; CHECK-LABEL: define <4 x i32> @test_mm_maskz_dpwsuds_epi32(
-; CHECK-SAME: i4 zeroext [[__U:%.*]], <4 x i32> [[__W:%.*]], <4 x i32> [[__A:%.*]], <4 x i32> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: i4 zeroext [[__U:%.*]], <4 x i32> [[__W:%.*]], <8 x i16> [[__A:%.*]], <8 x i16> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 24), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 8), align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 24), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i4, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <4 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> [[__W]], <4 x i32> [[__A]], <4 x i32> [[__B]])
+; CHECK-NEXT:    [[TMP18:%.*]] = icmp ne <8 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <8 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ne <8 x i16> [[__A]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ne <8 x i16> [[__B]], zeroinitializer
+; CHECK-NEXT:    [[TMP22:%.*]] = and <8 x i1> [[TMP18]], [[TMP19]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <8 x i1> [[TMP20]], [[TMP19]]
+; CHECK-NEXT:    [[TMP11:%.*]] = and <8 x i1> [[TMP18]], [[TMP21]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <8 x i1> [[TMP22]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = or <8 x i1> [[TMP12]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = sext <8 x i1> [[TMP13]] to <8 x i16>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast <8 x i16> [[TMP14]] to <4 x i32>
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <4 x i32> [[TMP15]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = sext <4 x i1> [[TMP16]] to <4 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[TMP17]], [[TMP1]]
+; CHECK-NEXT:    [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> [[__W]], <8 x i16> [[__A]], <8 x i16> [[__B]])
 ; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i4 [[TMP4]] to <4 x i1>
 ; CHECK-NEXT:    [[BST:%.*]] = bitcast i4 [[__U]] to <4 x i1>
 ; CHECK-NEXT:    [[TMP6:%.*]] = select <4 x i1> [[BST]], <4 x i32> [[_MSPROP1]], <4 x i32> zeroinitializer
@@ -789,23 +813,35 @@ define <4 x i32> @test_mm_maskz_dpwsuds_epi32(i4 zeroext %__U, <4 x i32> %__W, <
 ; CHECK-NEXT:    store <4 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <4 x i32> [[RES]]
 ;
-  %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B)
+  %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %__W, <8 x i16> %__A, <8 x i16> %__B)
   %bst = bitcast i4 %__U to <4 x i1>
   %res = select <4 x i1> %bst, <4 x i32> %dpi, <4 x i32> zeroinitializer
   ret <4 x i32> %res
 }
 
-define <8 x i32> @test_mm256_maskz_dpwsuds_epi32(<8 x i32> %__W, i8 zeroext %__U, <8 x i32> %__A, <8 x i32> %__B) sanitize_memory {
+define <8 x i32> @test_mm256_maskz_dpwsuds_epi32(<8 x i32> %__W, i8 zeroext %__U, <16 x i16> %__A, <16 x i16> %__B) sanitize_memory {
 ; CHECK-LABEL: define <8 x i32> @test_mm256_maskz_dpwsuds_epi32(
-; CHECK-SAME: <8 x i32> [[__W:%.*]], i8 zeroext [[__U:%.*]], <8 x i32> [[__A:%.*]], <8 x i32> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: <8 x i32> [[__W:%.*]], i8 zeroext [[__U:%.*]], <16 x i16> [[__A:%.*]], <16 x i16> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i8, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <8 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> [[__W]], <8 x i32> [[__A]], <8 x i32> [[__B]])
+; CHECK-NEXT:    [[TMP18:%.*]] = icmp ne <16 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <16 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ne <16 x i16> [[__A]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ne <16 x i16> [[__B]], zeroinitializer
+; CHECK-NEXT:    [[TMP22:%.*]] = and <16 x i1> [[TMP18]], [[TMP19]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <16 x i1> [[TMP20]], [[TMP19]]
+; CHECK-NEXT:    [[TMP11:%.*]] = and <16 x i1> [[TMP18]], [[TMP21]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <16 x i1> [[TMP22]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = or <16 x i1> [[TMP12]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = sext <16 x i1> [[TMP13]] to <16 x i16>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast <16 x i16> [[TMP14]] to <8 x i32>
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <8 x i32> [[TMP15]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = sext <8 x i1> [[TMP16]] to <8 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[TMP17]], [[TMP1]]
+; CHECK-NEXT:    [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> [[__W]], <16 x i16> [[__A]], <16 x i16> [[__B]])
 ; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>
 ; CHECK-NEXT:    [[BST:%.*]] = bitcast i8 [[__U]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP6:%.*]] = select <8 x i1> [[BST]], <8 x i32> [[_MSPROP1]], <8 x i32> [[TMP1]]
@@ -817,23 +853,35 @@ define <8 x i32> @test_mm256_maskz_dpwsuds_epi32(<8 x i32> %__W, i8 zeroext %__U
 ; CHECK-NEXT:    store <8 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <8 x i32> [[RES]]
 ;
-  %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B)
+  %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %__W, <16 x i16> %__A, <16 x i16> %__B)
   %bst = bitcast i8 %__U to <8 x i1>
   %res = select <8 x i1> %bst, <8 x i32> %dpi, <8 x i32> %__W
   ret <8 x i32> %res
 }
 
-define <8 x i32> @test_mm256_mask_dpwsud_epi32(i8 zeroext %__U, <8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B) sanitize_memory {
+define <8 x i32> @test_mm256_mask_dpwsud_epi32(i8 zeroext %__U, <8 x i32> %__W, <16 x i16> %__A, <16 x i16> %__B) sanitize_memory {
 ; CHECK-LABEL: define <8 x i32> @test_mm256_mask_dpwsud_epi32(
-; CHECK-SAME: i8 zeroext [[__U:%.*]], <8 x i32> [[__W:%.*]], <8 x i32> [[__A:%.*]], <8 x i32> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: i8 zeroext [[__U:%.*]], <8 x i32> [[__W:%.*]], <16 x i16> [[__A:%.*]], <16 x i16> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 8), align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i8, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <8 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> [[__W]], <8 x i32> [[__A]], <8 x i32> [[__B]])
+; CHECK-NEXT:    [[TMP18:%.*]] = icmp ne <16 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <16 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ne <16 x i16> [[__A]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ne <16 x i16> [[__B]], zeroinitializer
+; CHECK-NEXT:    [[TMP22:%.*]] = and <16 x i1> [[TMP18]], [[TMP19]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <16 x i1> [[TMP20]], [[TMP19]]
+; CHECK-NEXT:    [[TMP11:%.*]] = and <16 x i1> [[TMP18]], [[TMP21]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <16 x i1> [[TMP22]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = or <16 x i1> [[TMP12]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = sext <16 x i1> [[TMP13]] to <16 x i16>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast <16 x i16> [[TMP14]] to <8 x i32>
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <8 x i32> [[TMP15]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = sext <8 x i1> [[TMP16]] to <8 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[TMP17]], [[TMP1]]
+; CHECK-NEXT:    [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> [[__W]], <16 x i16> [[__A]], <16 x i16> [[__B]])
 ; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>
 ; CHECK-NEXT:    [[BST:%.*]] = bitcast i8 [[__U]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP6:%.*]] = select <8 x i1> [[BST]], <8 x i32> [[_MSPROP1]], <8 x i32> zeroinitializer
@@ -845,28 +893,40 @@ define <8 x i32> @test_mm256_mask_dpwsud_epi32(i8 zeroext %__U, <8 x i32> %__W,
 ; CHECK-NEXT:    store <8 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <8 x i32> [[RES]]
 ;
-  %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B)
+  %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %__W, <16 x i16> %__A, <16 x i16> %__B)
   %bst = bitcast i8 %__U to <8 x i1>
   %res = select <8 x i1> %bst, <8 x i32> %dpi, <8 x i32> zeroinitializer
   ret <8 x i32> %res
 }
 
-declare <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32>, <4 x i32>, <4 x i32>)
-declare <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32>, <4 x i32>, <4 x i32>)
-declare <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32>, <8 x i32>, <8 x i32>)
-declare <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32>, <8 x i32>, <8 x i32>)
+declare <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32>, <8 x i16>, <8 x i16>)
+declare <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32>, <8 x i16>, <8 x i16>)
+declare <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32>, <16 x i16>, <16 x i16>)
+declare <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32>, <16 x i16>, <16 x i16>)
 
-define <4 x i32> @test_mm_mask_dpwusd_epi32(<4 x i32> %__W, i4 zeroext %__U, <4 x i32> %__A, <4 x i32> %__B) sanitize_memory {
+define <4 x i32> @test_mm_mask_dpwusd_epi32(<4 x i32> %__W, i4 zeroext %__U, <8 x i16> %__A, <8 x i16> %__B) sanitize_memory {
 ; CHECK-LABEL: define <4 x i32> @test_mm_mask_dpwusd_epi32(
-; CHECK-SAME: <4 x i32> [[__W:%.*]], i4 zeroext [[__U:%.*]], <4 x i32> [[__A:%.*]], <4 x i32> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: <4 x i32> [[__W:%.*]], i4 zeroext [[__U:%.*]], <8 x i16> [[__A:%.*]], <8 x i16> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 24), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 24), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i4, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <4 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> [[__W]], <4 x i32> [[__A]], <4 x i32> [[__B]])
+; CHECK-NEXT:    [[TMP18:%.*]] = icmp ne <8 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <8 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ne <8 x i16> [[__A]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ne <8 x i16> [[__B]], zeroinitializer
+; CHECK-NEXT:    [[TMP22:%.*]] = and <8 x i1> [[TMP18]], [[TMP19]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <8 x i1> [[TMP20]], [[TMP19]]
+; CHECK-NEXT:    [[TMP11:%.*]] = and <8 x i1> [[TMP18]], [[TMP21]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <8 x i1> [[TMP22]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = or <8 x i1> [[TMP12]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = sext <8 x i1> [[TMP13]] to <8 x i16>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast <8 x i16> [[TMP14]] to <4 x i32>
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <4 x i32> [[TMP15]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = sext <4 x i1> [[TMP16]] to <4 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[TMP17]], [[TMP1]]
+; CHECK-NEXT:    [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> [[__W]], <8 x i16> [[__A]], <8 x i16> [[__B]])
 ; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i4 [[TMP4]] to <4 x i1>
 ; CHECK-NEXT:    [[BST:%.*]] = bitcast i4 [[__U]] to <4 x i1>
 ; CHECK-NEXT:    [[TMP6:%.*]] = select <4 x i1> [[BST]], <4 x i32> [[_MSPROP1]], <4 x i32> [[TMP1]]
@@ -878,23 +938,35 @@ define <4 x i32> @test_mm_mask_dpwusd_epi32(<4 x i32> %__W, i4 zeroext %__U, <4
 ; CHECK-NEXT:    store <4 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <4 x i32> [[RES]]
 ;
-  %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B)
+  %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> %__W, <8 x i16> %__A, <8 x i16> %__B)
   %bst = bitcast i4 %__U to <4 x i1>
   %res = select <4 x i1> %bst, <4 x i32> %dpi, <4 x i32> %__W
   ret <4 x i32> %res
 }
 
-define <4 x i32> @test_mm_maskz_dpwusds_epi32(i4 zeroext %__U, <4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B) sanitize_memory {
+define <4 x i32> @test_mm_maskz_dpwusds_epi32(i4 zeroext %__U, <4 x i32> %__W, <8 x i16> %__A, <8 x i16> %__B) sanitize_memory {
 ; CHECK-LABEL: define <4 x i32> @test_mm_maskz_dpwusds_epi32(
-; CHECK-SAME: i4 zeroext [[__U:%.*]], <4 x i32> [[__W:%.*]], <4 x i32> [[__A:%.*]], <4 x i32> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: i4 zeroext [[__U:%.*]], <4 x i32> [[__W:%.*]], <8 x i16> [[__A:%.*]], <8 x i16> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 24), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 8), align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 24), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i4, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <4 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> [[__W]], <4 x i32> [[__A]], <4 x i32> [[__B]])
+; CHECK-NEXT:    [[TMP18:%.*]] = icmp ne <8 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <8 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ne <8 x i16> [[__A]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ne <8 x i16> [[__B]], zeroinitializer
+; CHECK-NEXT:    [[TMP22:%.*]] = and <8 x i1> [[TMP18]], [[TMP19]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <8 x i1> [[TMP20]], [[TMP19]]
+; CHECK-NEXT:    [[TMP11:%.*]] = and <8 x i1> [[TMP18]], [[TMP21]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <8 x i1> [[TMP22]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = or <8 x i1> [[TMP12]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = sext <8 x i1> [[TMP13]] to <8 x i16>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast <8 x i16> [[TMP14]] to <4 x i32>
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <4 x i32> [[TMP15]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = sext <4 x i1> [[TMP16]] to <4 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[TMP17]], [[TMP1]]
+; CHECK-NEXT:    [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> [[__W]], <8 x i16> [[__A]], <8 x i16> [[__B]])
 ; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i4 [[TMP4]] to <4 x i1>
 ; CHECK-NEXT:    [[BST:%.*]] = bitcast i4 [[__U]] to <4 x i1>
 ; CHECK-NEXT:    [[TMP6:%.*]] = select <4 x i1> [[BST]], <4 x i32> [[_MSPROP1]], <4 x i32> zeroinitializer
@@ -906,23 +978,35 @@ define <4 x i32> @test_mm_maskz_dpwusds_epi32(i4 zeroext %__U, <4 x i32> %__W, <
 ; CHECK-NEXT:    store <4 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <4 x i32> [[RES]]
 ;
-  %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B)
+  %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> %__W, <8 x i16> %__A, <8 x i16> %__B)
   %bst = bitcast i4 %__U to <4 x i1>
   %res = select <4 x i1> %bst, <4 x i32> %dpi, <4 x i32> zeroinitializer
   ret <4 x i32> %res
 }
 
-define <8 x i32> @test_mm256_maskz_dpwusds_epi32(<8 x i32> %__W, i8 zeroext %__U, <8 x i32> %__A, <8 x i32> %__B) sanitize_memory {
+define <8 x i32> @test_mm256_maskz_dpwusds_epi32(<8 x i32> %__W, i8 zeroext %__U, <16 x i16> %__A, <16 x i16> %__B) sanitize_memory {
 ; CHECK-LABEL: define <8 x i32> @test_mm256_maskz_dpwusds_epi32(
-; CHECK-SAME: <8 x i32> [[__W:%.*]], i8 zeroext [[__U:%.*]], <8 x i32> [[__A:%.*]], <8 x i32> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: <8 x i32> [[__W:%.*]], i8 zeroext [[__U:%.*]], <16 x i16> [[__A:%.*]], <16 x i16> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i8, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <8 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32> [[__W]], <8 x i32> [[__A]], <8 x i32> [[__B]])
+; CHECK-NEXT:    [[TMP18:%.*]] = icmp ne <16 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <16 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ne <16 x i16> [[__A]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ne <16 x i16> [[__B]], zeroinitializer
+; CHECK-NEXT:    [[TMP22:%.*]] = and <16 x i1> [[TMP18]], [[TMP19]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <16 x i1> [[TMP20]], [[TMP19]]
+; CHECK-NEXT:    [[TMP11:%.*]] = and <16 x i1> [[TMP18]], [[TMP21]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <16 x i1> [[TMP22]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = or <16 x i1> [[TMP12]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = sext <16 x i1> [[TMP13]] to <16 x i16>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast <16 x i16> [[TMP14]] to <8 x i32>
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <8 x i32> [[TMP15]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = sext <8 x i1> [[TMP16]] to <8 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[TMP17]], [[TMP1]]
+; CHECK-NEXT:    [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32> [[__W]], <16 x i16> [[__A]], <16 x i16> [[__B]])
 ; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>
 ; CHECK-NEXT:    [[BST:%.*]] = bitcast i8 [[__U]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP6:%.*]] = select <8 x i1> [[BST]], <8 x i32> [[_MSPROP1]], <8 x i32> [[TMP1]]
@@ -934,23 +1018,35 @@ define <8 x i32> @test_mm256_maskz_dpwusds_epi32(<8 x i32> %__W, i8 zeroext %__U
 ; CHECK-NEXT:    store <8 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <8 x i32> [[RES]]
 ;
-  %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B)
+  %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32> %__W, <16 x i16> %__A, <16 x i16> %__B)
   %bst = bitcast i8 %__U to <8 x i1>
   %res = select <8 x i1> %bst, <8 x i32> %dpi, <8 x i32> %__W
   ret <8 x i32> %res
 }
 
-define <8 x i32> @test_mm256_mask_dpwusd_epi32(i8 zeroext %__U, <8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B) sanitize_memory {
+define <8 x i32> @test_mm256_mask_dpwusd_epi32(i8 zeroext %__U, <8 x i32> %__W, <16 x i16> %__A, <16 x i16> %__B) sanitize_memory {
 ; CHECK-LABEL: define <8 x i32> @test_mm256_mask_dpwusd_epi32(
-; CHECK-SAME: i8 zeroext [[__U:%.*]], <8 x i32> [[__W:%.*]], <8 x i32> [[__A:%.*]], <8 x i32> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: i8 zeroext [[__U:%.*]], <8 x i32> [[__W:%.*]], <16 x i16> [[__A:%.*]], <16 x i16> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 8), align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i8, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <8 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> [[__W]], <8 x i32> [[__A]], <8 x i32> [[__B]])
+; CHECK-NEXT:    [[TMP18:%.*]] = icmp ne <16 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <16 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ne <16 x i16> [[__A]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ne <16 x i16> [[__B]], zeroinitializer
+; CHECK-NEXT:    [[TMP22:%.*]] = and <16 x i1> [[TMP18]], [[TMP19]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <16 x i1> [[TMP20]], [[TMP19]]
+; CHECK-NEXT:    [[TMP11:%.*]] = and <16 x i1> [[TMP18]], [[TMP21]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <16 x i1> [[TMP22]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = or <16 x i1> [[TMP12]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = sext <16 x i1> [[TMP13]] to <16 x i16>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast <16 x i16> [[TMP14]] to <8 x i32>
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <8 x i32> [[TMP15]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = sext <8 x i1> [[TMP16]] to <8 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[TMP17]], [[TMP1]]
+; CHECK-NEXT:    [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> [[__W]], <16 x i16> [[__A]], <16 x i16> [[__B]])
 ; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>
 ; CHECK-NEXT:    [[BST:%.*]] = bitcast i8 [[__U]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP6:%.*]] = select <8 x i1> [[BST]], <8 x i32> [[_MSPROP1]], <8 x i32> zeroinitializer
@@ -962,28 +1058,40 @@ define <8 x i32> @test_mm256_mask_dpwusd_epi32(i8 zeroext %__U, <8 x i32> %__W,
 ; CHECK-NEXT:    store <8 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <8 x i32> [[RES]]
 ;
-  %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B)
+  %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> %__W, <16 x i16> %__A, <16 x i16> %__B)
   %bst = bitcast i8 %__U to <8 x i1>
   %res = select <8 x i1> %bst, <8 x i32> %dpi, <8 x i32> zeroinitializer
   ret <8 x i32> %res
 }
 
-declare <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32>, <4 x i32>, <4 x i32>)
-declare <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32>, <4 x i32>, <4 x i32>)
-declare <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32>, <8 x i32>, <8 x i32>)
-declare <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32>, <8 x i32>, <8 x i32>)
+declare <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32>, <8 x i16>, <8 x i16>)
+declare <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32>, <8 x i16>, <8 x i16>)
+declare <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32>, <16 x i16>, <16 x i16>)
+declare <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32>, <16 x i16>, <16 x i16>)
 
-define <4 x i32> @test_mm_mask_dpwuud_epi32(<4 x i32> %__W, i4 zeroext %__U, <4 x i32> %__A, <4 x i32> %__B) sanitize_memory {
+define <4 x i32> @test_mm_mask_dpwuud_epi32(<4 x i32> %__W, i4 zeroext %__U, <8 x i16> %__A, <8 x i16> %__B) sanitize_memory {
 ; CHECK-LABEL: define <4 x i32> @test_mm_mask_dpwuud_epi32(
-; CHECK-SAME: <4 x i32> [[__W:%.*]], i4 zeroext [[__U:%.*]], <4 x i32> [[__A:%.*]], <4 x i32> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: <4 x i32> [[__W:%.*]], i4 zeroext [[__U:%.*]], <8 x i16> [[__A:%.*]], <8 x i16> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 24), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 24), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i4, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <4 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32> [[__W]], <4 x i32> [[__A]], <4 x i32> [[__B]])
+; CHECK-NEXT:    [[TMP18:%.*]] = icmp ne <8 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <8 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ne <8 x i16> [[__A]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ne <8 x i16> [[__B]], zeroinitializer
+; CHECK-NEXT:    [[TMP22:%.*]] = and <8 x i1> [[TMP18]], [[TMP19]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <8 x i1> [[TMP20]], [[TMP19]]
+; CHECK-NEXT:    [[TMP11:%.*]] = and <8 x i1> [[TMP18]], [[TMP21]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <8 x i1> [[TMP22]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = or <8 x i1> [[TMP12]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = sext <8 x i1> [[TMP13]] to <8 x i16>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast <8 x i16> [[TMP14]] to <4 x i32>
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <4 x i32> [[TMP15]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = sext <4 x i1> [[TMP16]] to <4 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[TMP17]], [[TMP1]]
+; CHECK-NEXT:    [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32> [[__W]], <8 x i16> [[__A]], <8 x i16> [[__B]])
 ; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i4 [[TMP4]] to <4 x i1>
 ; CHECK-NEXT:    [[BST:%.*]] = bitcast i4 [[__U]] to <4 x i1>
 ; CHECK-NEXT:    [[TMP6:%.*]] = select <4 x i1> [[BST]], <4 x i32> [[_MSPROP1]], <4 x i32> [[TMP1]]
@@ -995,23 +1103,35 @@ define <4 x i32> @test_mm_mask_dpwuud_epi32(<4 x i32> %__W, i4 zeroext %__U, <4
 ; CHECK-NEXT:    store <4 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <4 x i32> [[RES]]
 ;
-  %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B)
+  %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32> %__W, <8 x i16> %__A, <8 x i16> %__B)
   %bst = bitcast i4 %__U to <4 x i1>
   %res = select <4 x i1> %bst, <4 x i32> %dpi, <4 x i32> %__W
   ret <4 x i32> %res
 }
 
-define <4 x i32> @test_mm_maskz_dpwuuds_epi32(i4 zeroext %__U, <4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B) sanitize_memory {
+define <4 x i32> @test_mm_maskz_dpwuuds_epi32(i4 zeroext %__U, <4 x i32> %__W, <8 x i16> %__A, <8 x i16> %__B) sanitize_memory {
 ; CHECK-LABEL: define <4 x i32> @test_mm_maskz_dpwuuds_epi32(
-; CHECK-SAME: i4 zeroext [[__U:%.*]], <4 x i32> [[__W:%.*]], <4 x i32> [[__A:%.*]], <4 x i32> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: i4 zeroext [[__U:%.*]], <4 x i32> [[__W:%.*]], <8 x i16> [[__A:%.*]], <8 x i16> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 24), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 8), align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 24), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i4, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <4 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> [[__W]], <4 x i32> [[__A]], <4 x i32> [[__B]])
+; CHECK-NEXT:    [[TMP18:%.*]] = icmp ne <8 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <8 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ne <8 x i16> [[__A]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ne <8 x i16> [[__B]], zeroinitializer
+; CHECK-NEXT:    [[TMP22:%.*]] = and <8 x i1> [[TMP18]], [[TMP19]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <8 x i1> [[TMP20]], [[TMP19]]
+; CHECK-NEXT:    [[TMP11:%.*]] = and <8 x i1> [[TMP18]], [[TMP21]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <8 x i1> [[TMP22]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = or <8 x i1> [[TMP12]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = sext <8 x i1> [[TMP13]] to <8 x i16>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast <8 x i16> [[TMP14]] to <4 x i32>
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <4 x i32> [[TMP15]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = sext <4 x i1> [[TMP16]] to <4 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[TMP17]], [[TMP1]]
+; CHECK-NEXT:    [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> [[__W]], <8 x i16> [[__A]], <8 x i16> [[__B]])
 ; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i4 [[TMP4]] to <4 x i1>
 ; CHECK-NEXT:    [[BST:%.*]] = bitcast i4 [[__U]] to <4 x i1>
 ; CHECK-NEXT:    [[TMP6:%.*]] = select <4 x i1> [[BST]], <4 x i32> [[_MSPROP1]], <4 x i32> zeroinitializer
@@ -1023,23 +1143,35 @@ define <4 x i32> @test_mm_maskz_dpwuuds_epi32(i4 zeroext %__U, <4 x i32> %__W, <
 ; CHECK-NEXT:    store <4 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <4 x i32> [[RES]]
 ;
-  %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B)
+  %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> %__W, <8 x i16> %__A, <8 x i16> %__B)
   %bst = bitcast i4 %__U to <4 x i1>
   %res = select <4 x i1> %bst, <4 x i32> %dpi, <4 x i32> zeroinitializer
   ret <4 x i32> %res
 }
 
-define <8 x i32> @test_mm256_maskz_dpwuuds_epi32(<8 x i32> %__W, i8 zeroext %__U, <8 x i32> %__A, <8 x i32> %__B) sanitize_memory {
+define <8 x i32> @test_mm256_maskz_dpwuuds_epi32(<8 x i32> %__W, i8 zeroext %__U, <16 x i16> %__A, <16 x i16> %__B) sanitize_memory {
 ; CHECK-LABEL: define <8 x i32> @test_mm256_maskz_dpwuuds_epi32(
-; CHECK-SAME: <8 x i32> [[__W:%.*]], i8 zeroext [[__U:%.*]], <8 x i32> [[__A:%.*]], <8 x i32> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: <8 x i32> [[__W:%.*]], i8 zeroext [[__U:%.*]], <16 x i16> [[__A:%.*]], <16 x i16> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i8, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <8 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> [[__W]], <8 x i32> [[__A]], <8 x i32> [[__B]])
+; CHECK-NEXT:    [[TMP18:%.*]] = icmp ne <16 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <16 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ne <16 x i16> [[__A]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ne <16 x i16> [[__B]], zeroinitializer
+; CHECK-NEXT:    [[TMP22:%.*]] = and <16 x i1> [[TMP18]], [[TMP19]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <16 x i1> [[TMP20]], [[TMP19]]
+; CHECK-NEXT:    [[TMP11:%.*]] = and <16 x i1> [[TMP18]], [[TMP21]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <16 x i1> [[TMP22]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = or <16 x i1> [[TMP12]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = sext <16 x i1> [[TMP13]] to <16 x i16>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast <16 x i16> [[TMP14]] to <8 x i32>
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <8 x i32> [[TMP15]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = sext <8 x i1> [[TMP16]] to <8 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[TMP17]], [[TMP1]]
+; CHECK-NEXT:    [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> [[__W]], <16 x i16> [[__A]], <16 x i16> [[__B]])
 ; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>
 ; CHECK-NEXT:    [[BST:%.*]] = bitcast i8 [[__U]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP6:%.*]] = select <8 x i1> [[BST]], <8 x i32> [[_MSPROP1]], <8 x i32> [[TMP1]]
@@ -1051,23 +1183,35 @@ define <8 x i32> @test_mm256_maskz_dpwuuds_epi32(<8 x i32> %__W, i8 zeroext %__U
 ; CHECK-NEXT:    store <8 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <8 x i32> [[RES]]
 ;
-  %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B)
+  %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> %__W, <16 x i16> %__A, <16 x i16> %__B)
   %bst = bitcast i8 %__U to <8 x i1>
   %res = select <8 x i1> %bst, <8 x i32> %dpi, <8 x i32> %__W
   ret <8 x i32> %res
 }
 
-define <8 x i32> @test_mm256_mask_dpwuud_epi32(i8 zeroext %__U, <8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B) sanitize_memory {
+define <8 x i32> @test_mm256_mask_dpwuud_epi32(i8 zeroext %__U, <8 x i32> %__W, <16 x i16> %__A, <16 x i16> %__B) sanitize_memory {
 ; CHECK-LABEL: define <8 x i32> @test_mm256_mask_dpwuud_epi32(
-; CHECK-SAME: i8 zeroext [[__U:%.*]], <8 x i32> [[__W:%.*]], <8 x i32> [[__A:%.*]], <8 x i32> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: i8 zeroext [[__U:%.*]], <8 x i32> [[__W:%.*]], <16 x i16> [[__A:%.*]], <16 x i16> [[__B:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 8), align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 40), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 72), align 8
 ; CHECK-NEXT:    [[TMP4:%.*]] = load i8, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <8 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> [[__W]], <8 x i32> [[__A]], <8 x i32> [[__B]])
+; CHECK-NEXT:    [[TMP18:%.*]] = icmp ne <16 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = icmp ne <16 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp ne <16 x i16> [[__A]], zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp ne <16 x i16> [[__B]], zeroinitializer
+; CHECK-NEXT:    [[TMP22:%.*]] = and <16 x i1> [[TMP18]], [[TMP19]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <16 x i1> [[TMP20]], [[TMP19]]
+; CHECK-NEXT:    [[TMP11:%.*]] = and <16 x i1> [[TMP18]], [[TMP21]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <16 x i1> [[TMP22]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = or <16 x i1> [[TMP12]], [[TMP11]]
+; CHECK-NEXT:    [[TMP14:%.*]] = sext <16 x i1> [[TMP13]] to <16 x i16>
+; CHECK-NEXT:    [[TMP15:%.*]] = bitcast <16 x i16> [[TMP14]] to <8 x i32>
+; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne <8 x i32> [[TMP15]], zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = sext <8 x i1> [[TMP16]] to <8 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[TMP17]], [[TMP1]]
+; CHECK-NEXT:    [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> [[__W]], <16 x i16> [[__A]], <16 x i16> [[__B]])
 ; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i8 [[TMP4]] to <8 x i1>
 ; CHECK-NEXT:    [[BST:%.*]] = bitcast i8 [[__U]] to <8 x i1>
 ; CHECK-NEXT:    [[TMP6:%.*]] = select <8 x i1> [[BST]], <8 x i32> [[_MSPROP1]], <8 x i32> zeroinitializer
@@ -1079,16 +1223,16 @@ define <8 x i32> @test_mm256_mask_dpwuud_epi32(i8 zeroext %__U, <8 x i32> %__W,
 ; CHECK-NEXT:    store <8 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <8 x i32> [[RES]]
 ;
-  %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B)
+  %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> %__W, <16 x i16> %__A, <16 x i16> %__B)
   %bst = bitcast i8 %__U to <8 x i1>
   %res = select <8 x i1> %bst, <8 x i32> %dpi, <8 x i32> zeroinitializer
   ret <8 x i32> %res
 }
 
-declare <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32>, <4 x i32>, <4 x i32>)
-declare <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32>, <4 x i32>, <4 x i32>)
-declare <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32>, <8 x i32>, <8 x i32>)
-declare <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32>, <8 x i32>, <8 x i32>)
+declare <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32>, <8 x i16>, <8 x i16>)
+declare <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32>, <8 x i16>, <8 x i16>)
+declare <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32>, <16 x i16>, <16 x i16>)
+declare <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32>, <16 x i16>, <16 x i16>)
 
 
 define { <8 x i16>, <8 x i16>, <8 x i16> } @test_mask_mpsadbw_128(<16 x i8> %x0, <16 x i8> %x1, <8 x i16> %x3, i8 %x4) sanitize_memory {
diff --git a/llvm/test/Instrumentation/MemorySanitizer/X86/avxvnniint16-intrinsics.ll b/llvm/test/Instrumentation/MemorySanitizer/X86/avxvnniint16-intrinsics.ll
index 707b46bb8686e..fd9e0b953d002 100644
--- a/llvm/test/Instrumentation/MemorySanitizer/X86/avxvnniint16-intrinsics.ll
+++ b/llvm/test/Instrumentation/MemorySanitizer/X86/avxvnniint16-intrinsics.ll
@@ -22,218 +22,362 @@
 target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-unknown-linux-gnu"
 
-define <4 x i32> @test_int_x86_avx2_vpdpwsud_128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C) sanitize_memory {
+define <4 x i32> @test_int_x86_avx2_vpdpwsud_128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C) sanitize_memory {
 ; CHECK-LABEL: define <4 x i32> @test_int_x86_avx2_vpdpwsud_128(
-; CHECK-SAME: <4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]], <4 x i32> [[C:%.*]]) #[[ATTR0:[0-9]+]] {
-; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
+; CHECK-SAME: <4 x i32> [[A:%.*]], <8 x i16> [[B:%.*]], <8 x i16> [[C:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
+; CHECK-NEXT:    [[TMP18:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <4 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[RET:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> [[A]], <4 x i32> [[B]], <4 x i32> [[C]])
-; CHECK-NEXT:    store <4 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne <8 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne <8 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne <8 x i16> [[B]], zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp ne <8 x i16> [[C]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = and <8 x i1> [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP9:%.*]] = and <8 x i1> [[TMP6]], [[TMP5]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <8 x i1> [[TMP4]], [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <8 x i1> [[TMP8]], [[TMP9]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <8 x i1> [[TMP11]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = sext <8 x i1> [[TMP12]] to <8 x i16>
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast <8 x i16> [[TMP13]] to <4 x i32>
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp ne <4 x i32> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[TMP16:%.*]] = sext <4 x i1> [[TMP15]] to <4 x i32>
+; CHECK-NEXT:    [[TMP17:%.*]] = or <4 x i32> [[TMP16]], [[TMP18]]
+; CHECK-NEXT:    [[RET:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> [[A]], <8 x i16> [[B]], <8 x i16> [[C]])
+; CHECK-NEXT:    store <4 x i32> [[TMP17]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <4 x i32> [[RET]]
 ;
-  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
   ret <4 x i32> %ret
 }
-declare <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+declare <4 x i32> @llvm.x86.avx2.vpdpwsud.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
 
-define <8 x i32> @test_int_x86_avx2_vpdpwsud_256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C) sanitize_memory {
+define <8 x i32> @test_int_x86_avx2_vpdpwsud_256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C) sanitize_memory {
 ; CHECK-LABEL: define <8 x i32> @test_int_x86_avx2_vpdpwsud_256(
-; CHECK-SAME: <8 x i32> [[A:%.*]], <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:    [[TMP1:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
+; CHECK-SAME: <8 x i32> [[A:%.*]], <16 x i16> [[B:%.*]], <16 x i16> [[C:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
+; CHECK-NEXT:    [[TMP18:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <8 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[RET:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> [[C]])
-; CHECK-NEXT:    store <8 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne <16 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne <16 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne <16 x i16> [[B]], zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp ne <16 x i16> [[C]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = and <16 x i1> [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP9:%.*]] = and <16 x i1> [[TMP6]], [[TMP5]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <16 x i1> [[TMP4]], [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <16 x i1> [[TMP8]], [[TMP9]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <16 x i1> [[TMP11]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = sext <16 x i1> [[TMP12]] to <16 x i16>
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast <16 x i16> [[TMP13]] to <8 x i32>
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp ne <8 x i32> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[TMP16:%.*]] = sext <8 x i1> [[TMP15]] to <8 x i32>
+; CHECK-NEXT:    [[TMP17:%.*]] = or <8 x i32> [[TMP16]], [[TMP18]]
+; CHECK-NEXT:    [[RET:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> [[A]], <16 x i16> [[B]], <16 x i16> [[C]])
+; CHECK-NEXT:    store <8 x i32> [[TMP17]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <8 x i32> [[RET]]
 ;
-  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
   ret <8 x i32> %ret
 }
-declare <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+declare <8 x i32> @llvm.x86.avx2.vpdpwsud.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
 
-define <4 x i32> @test_int_x86_avx2_vpdpwsuds_128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C) sanitize_memory {
+define <4 x i32> @test_int_x86_avx2_vpdpwsuds_128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C) sanitize_memory {
 ; CHECK-LABEL: define <4 x i32> @test_int_x86_avx2_vpdpwsuds_128(
-; CHECK-SAME: <4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]], <4 x i32> [[C:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
+; CHECK-SAME: <4 x i32> [[A:%.*]], <8 x i16> [[B:%.*]], <8 x i16> [[C:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
+; CHECK-NEXT:    [[TMP18:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <4 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[RET:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> [[A]], <4 x i32> [[B]], <4 x i32> [[C]])
-; CHECK-NEXT:    store <4 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne <8 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne <8 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne <8 x i16> [[B]], zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp ne <8 x i16> [[C]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = and <8 x i1> [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP9:%.*]] = and <8 x i1> [[TMP6]], [[TMP5]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <8 x i1> [[TMP4]], [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <8 x i1> [[TMP8]], [[TMP9]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <8 x i1> [[TMP11]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = sext <8 x i1> [[TMP12]] to <8 x i16>
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast <8 x i16> [[TMP13]] to <4 x i32>
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp ne <4 x i32> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[TMP16:%.*]] = sext <4 x i1> [[TMP15]] to <4 x i32>
+; CHECK-NEXT:    [[TMP17:%.*]] = or <4 x i32> [[TMP16]], [[TMP18]]
+; CHECK-NEXT:    [[RET:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> [[A]], <8 x i16> [[B]], <8 x i16> [[C]])
+; CHECK-NEXT:    store <4 x i32> [[TMP17]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <4 x i32> [[RET]]
 ;
-  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
   ret <4 x i32> %ret
 }
-declare <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+declare <4 x i32> @llvm.x86.avx2.vpdpwsuds.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
 
-define <8 x i32> @test_int_x86_avx2_vpdpwsuds_256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C) sanitize_memory {
+define <8 x i32> @test_int_x86_avx2_vpdpwsuds_256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C) sanitize_memory {
 ; CHECK-LABEL: define <8 x i32> @test_int_x86_avx2_vpdpwsuds_256(
-; CHECK-SAME: <8 x i32> [[A:%.*]], <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT:    [[TMP1:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
+; CHECK-SAME: <8 x i32> [[A:%.*]], <16 x i16> [[B:%.*]], <16 x i16> [[C:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
+; CHECK-NEXT:    [[TMP18:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <8 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[RET:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> [[C]])
-; CHECK-NEXT:    store <8 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne <16 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne <16 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne <16 x i16> [[B]], zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp ne <16 x i16> [[C]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = and <16 x i1> [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP9:%.*]] = and <16 x i1> [[TMP6]], [[TMP5]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <16 x i1> [[TMP4]], [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <16 x i1> [[TMP8]], [[TMP9]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <16 x i1> [[TMP11]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = sext <16 x i1> [[TMP12]] to <16 x i16>
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast <16 x i16> [[TMP13]] to <8 x i32>
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp ne <8 x i32> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[TMP16:%.*]] = sext <8 x i1> [[TMP15]] to <8 x i32>
+; CHECK-NEXT:    [[TMP17:%.*]] = or <8 x i32> [[TMP16]], [[TMP18]]
+; CHECK-NEXT:    [[RET:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> [[A]], <16 x i16> [[B]], <16 x i16> [[C]])
+; CHECK-NEXT:    store <8 x i32> [[TMP17]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <8 x i32> [[RET]]
 ;
-  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
   ret <8 x i32> %ret
 }
-declare <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+declare <8 x i32> @llvm.x86.avx2.vpdpwsuds.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
 
-define <4 x i32> @test_int_x86_avx2_vpdpwusd_128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C) sanitize_memory {
+define <4 x i32> @test_int_x86_avx2_vpdpwusd_128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C) sanitize_memory {
 ; CHECK-LABEL: define <4 x i32> @test_int_x86_avx2_vpdpwusd_128(
-; CHECK-SAME: <4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]], <4 x i32> [[C:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: <4 x i32> [[A:%.*]], <8 x i16> [[B:%.*]], <8 x i16> [[C:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <4 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[RET:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> [[A]], <4 x i32> [[B]], <4 x i32> [[C]])
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne <8 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne <8 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne <8 x i16> [[B]], zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp ne <8 x i16> [[C]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = and <8 x i1> [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP9:%.*]] = and <8 x i1> [[TMP6]], [[TMP5]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <8 x i1> [[TMP4]], [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <8 x i1> [[TMP8]], [[TMP9]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <8 x i1> [[TMP11]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = sext <8 x i1> [[TMP12]] to <8 x i16>
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast <8 x i16> [[TMP13]] to <4 x i32>
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp ne <4 x i32> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[TMP16:%.*]] = sext <4 x i1> [[TMP15]] to <4 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[TMP16]], [[TMP1]]
+; CHECK-NEXT:    [[RET:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> [[A]], <8 x i16> [[B]], <8 x i16> [[C]])
 ; CHECK-NEXT:    store <4 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <4 x i32> [[RET]]
 ;
-  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
   ret <4 x i32> %ret
 }
-declare <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+declare <4 x i32> @llvm.x86.avx2.vpdpwusd.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
 
-define <8 x i32> @test_int_x86_avx2_vpdpwusd_256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C) sanitize_memory {
+define <8 x i32> @test_int_x86_avx2_vpdpwusd_256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C) sanitize_memory {
 ; CHECK-LABEL: define <8 x i32> @test_int_x86_avx2_vpdpwusd_256(
-; CHECK-SAME: <8 x i32> [[A:%.*]], <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: <8 x i32> [[A:%.*]], <16 x i16> [[B:%.*]], <16 x i16> [[C:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <8 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[RET:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> [[C]])
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne <16 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne <16 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne <16 x i16> [[B]], zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp ne <16 x i16> [[C]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = and <16 x i1> [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP9:%.*]] = and <16 x i1> [[TMP6]], [[TMP5]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <16 x i1> [[TMP4]], [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <16 x i1> [[TMP8]], [[TMP9]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <16 x i1> [[TMP11]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = sext <16 x i1> [[TMP12]] to <16 x i16>
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast <16 x i16> [[TMP13]] to <8 x i32>
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp ne <8 x i32> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[TMP16:%.*]] = sext <8 x i1> [[TMP15]] to <8 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[TMP16]], [[TMP1]]
+; CHECK-NEXT:    [[RET:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> [[A]], <16 x i16> [[B]], <16 x i16> [[C]])
 ; CHECK-NEXT:    store <8 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <8 x i32> [[RET]]
 ;
-  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
   ret <8 x i32> %ret
 }
-declare <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+declare <8 x i32> @llvm.x86.avx2.vpdpwusd.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
 
-define <4 x i32> @test_int_x86_avx2_vpdpwusds_128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C) sanitize_memory {
+define <4 x i32> @test_int_x86_avx2_vpdpwusds_128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C) sanitize_memory {
 ; CHECK-LABEL: define <4 x i32> @test_int_x86_avx2_vpdpwusds_128(
-; CHECK-SAME: <4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]], <4 x i32> [[C:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: <4 x i32> [[A:%.*]], <8 x i16> [[B:%.*]], <8 x i16> [[C:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <4 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[RET:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> [[A]], <4 x i32> [[B]], <4 x i32> [[C]])
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne <8 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne <8 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne <8 x i16> [[B]], zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp ne <8 x i16> [[C]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = and <8 x i1> [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP9:%.*]] = and <8 x i1> [[TMP6]], [[TMP5]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <8 x i1> [[TMP4]], [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <8 x i1> [[TMP8]], [[TMP9]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <8 x i1> [[TMP11]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = sext <8 x i1> [[TMP12]] to <8 x i16>
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast <8 x i16> [[TMP13]] to <4 x i32>
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp ne <4 x i32> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[TMP16:%.*]] = sext <4 x i1> [[TMP15]] to <4 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[TMP16]], [[TMP1]]
+; CHECK-NEXT:    [[RET:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> [[A]], <8 x i16> [[B]], <8 x i16> [[C]])
 ; CHECK-NEXT:    store <4 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <4 x i32> [[RET]]
 ;
-  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
   ret <4 x i32> %ret
 }
-declare <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+declare <4 x i32> @llvm.x86.avx2.vpdpwusds.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
 
-define <8 x i32> @test_int_x86_avx2_vpdpwusds_256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C) sanitize_memory {
+define <8 x i32> @test_int_x86_avx2_vpdpwusds_256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C) sanitize_memory {
 ; CHECK-LABEL: define <8 x i32> @test_int_x86_avx2_vpdpwusds_256(
-; CHECK-SAME: <8 x i32> [[A:%.*]], <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: <8 x i32> [[A:%.*]], <16 x i16> [[B:%.*]], <16 x i16> [[C:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <8 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[RET:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> [[C]])
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne <16 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne <16 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne <16 x i16> [[B]], zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp ne <16 x i16> [[C]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = and <16 x i1> [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP9:%.*]] = and <16 x i1> [[TMP6]], [[TMP5]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <16 x i1> [[TMP4]], [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <16 x i1> [[TMP8]], [[TMP9]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <16 x i1> [[TMP11]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = sext <16 x i1> [[TMP12]] to <16 x i16>
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast <16 x i16> [[TMP13]] to <8 x i32>
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp ne <8 x i32> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[TMP16:%.*]] = sext <8 x i1> [[TMP15]] to <8 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[TMP16]], [[TMP1]]
+; CHECK-NEXT:    [[RET:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32> [[A]], <16 x i16> [[B]], <16 x i16> [[C]])
 ; CHECK-NEXT:    store <8 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <8 x i32> [[RET]]
 ;
-  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
   ret <8 x i32> %ret
 }
-declare <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+declare <8 x i32> @llvm.x86.avx2.vpdpwusds.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
 
-define <4 x i32> @test_int_x86_avx2_vpdpwuud_128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C) sanitize_memory {
+define <4 x i32> @test_int_x86_avx2_vpdpwuud_128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C) sanitize_memory {
 ; CHECK-LABEL: define <4 x i32> @test_int_x86_avx2_vpdpwuud_128(
-; CHECK-SAME: <4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]], <4 x i32> [[C:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: <4 x i32> [[A:%.*]], <8 x i16> [[B:%.*]], <8 x i16> [[C:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <4 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[RET:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32> [[A]], <4 x i32> [[B]], <4 x i32> [[C]])
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne <8 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne <8 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne <8 x i16> [[B]], zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp ne <8 x i16> [[C]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = and <8 x i1> [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP9:%.*]] = and <8 x i1> [[TMP6]], [[TMP5]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <8 x i1> [[TMP4]], [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <8 x i1> [[TMP8]], [[TMP9]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <8 x i1> [[TMP11]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = sext <8 x i1> [[TMP12]] to <8 x i16>
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast <8 x i16> [[TMP13]] to <4 x i32>
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp ne <4 x i32> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[TMP16:%.*]] = sext <4 x i1> [[TMP15]] to <4 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[TMP16]], [[TMP1]]
+; CHECK-NEXT:    [[RET:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32> [[A]], <8 x i16> [[B]], <8 x i16> [[C]])
 ; CHECK-NEXT:    store <4 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <4 x i32> [[RET]]
 ;
-  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
   ret <4 x i32> %ret
 }
-declare <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+declare <4 x i32> @llvm.x86.avx2.vpdpwuud.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
 
-define <8 x i32> @test_int_x86_avx2_vpdpwuud_256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C) sanitize_memory {
+define <8 x i32> @test_int_x86_avx2_vpdpwuud_256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C) sanitize_memory {
 ; CHECK-LABEL: define <8 x i32> @test_int_x86_avx2_vpdpwuud_256(
-; CHECK-SAME: <8 x i32> [[A:%.*]], <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: <8 x i32> [[A:%.*]], <16 x i16> [[B:%.*]], <16 x i16> [[C:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <8 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[RET:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> [[C]])
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne <16 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne <16 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne <16 x i16> [[B]], zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp ne <16 x i16> [[C]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = and <16 x i1> [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP9:%.*]] = and <16 x i1> [[TMP6]], [[TMP5]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <16 x i1> [[TMP4]], [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <16 x i1> [[TMP8]], [[TMP9]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <16 x i1> [[TMP11]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = sext <16 x i1> [[TMP12]] to <16 x i16>
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast <16 x i16> [[TMP13]] to <8 x i32>
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp ne <8 x i32> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[TMP16:%.*]] = sext <8 x i1> [[TMP15]] to <8 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[TMP16]], [[TMP1]]
+; CHECK-NEXT:    [[RET:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> [[A]], <16 x i16> [[B]], <16 x i16> [[C]])
 ; CHECK-NEXT:    store <8 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <8 x i32> [[RET]]
 ;
-  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
   ret <8 x i32> %ret
 }
-declare <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+declare <8 x i32> @llvm.x86.avx2.vpdpwuud.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
 
-define <4 x i32> @test_int_x86_avx2_vpdpwuuds_128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C) sanitize_memory {
+define <4 x i32> @test_int_x86_avx2_vpdpwuuds_128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C) sanitize_memory {
 ; CHECK-LABEL: define <4 x i32> @test_int_x86_avx2_vpdpwuuds_128(
-; CHECK-SAME: <4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]], <4 x i32> [[C:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: <4 x i32> [[A:%.*]], <8 x i16> [[B:%.*]], <8 x i16> [[C:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 16), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <4 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[RET:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> [[A]], <4 x i32> [[B]], <4 x i32> [[C]])
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne <8 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne <8 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne <8 x i16> [[B]], zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp ne <8 x i16> [[C]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = and <8 x i1> [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP9:%.*]] = and <8 x i1> [[TMP6]], [[TMP5]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <8 x i1> [[TMP4]], [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <8 x i1> [[TMP8]], [[TMP9]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <8 x i1> [[TMP11]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = sext <8 x i1> [[TMP12]] to <8 x i16>
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast <8 x i16> [[TMP13]] to <4 x i32>
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp ne <4 x i32> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[TMP16:%.*]] = sext <4 x i1> [[TMP15]] to <4 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <4 x i32> [[TMP16]], [[TMP1]]
+; CHECK-NEXT:    [[RET:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> [[A]], <8 x i16> [[B]], <8 x i16> [[C]])
 ; CHECK-NEXT:    store <4 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <4 x i32> [[RET]]
 ;
-  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+  %ret = call <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
   ret <4 x i32> %ret
 }
-declare <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> %A, <4 x i32> %B, <4 x i32> %C)
+declare <4 x i32> @llvm.x86.avx2.vpdpwuuds.128(<4 x i32> %A, <8 x i16> %B, <8 x i16> %C)
 
-define <8 x i32> @test_int_x86_avx2_vpdpwuuds_256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C) sanitize_memory {
+define <8 x i32> @test_int_x86_avx2_vpdpwuuds_256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C) sanitize_memory {
 ; CHECK-LABEL: define <8 x i32> @test_int_x86_avx2_vpdpwuuds_256(
-; CHECK-SAME: <8 x i32> [[A:%.*]], <8 x i32> [[B:%.*]], <8 x i32> [[C:%.*]]) #[[ATTR0]] {
+; CHECK-SAME: <8 x i32> [[A:%.*]], <16 x i16> [[B:%.*]], <16 x i16> [[C:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[TMP3:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i16>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8
-; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 32), align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x i32>, ptr getelementptr (i8, ptr @__msan_param_tls, i64 64), align 8
 ; CHECK-NEXT:    call void @llvm.donothing()
-; CHECK-NEXT:    [[_MSPROP:%.*]] = or <8 x i32> [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[_MSPROP]], [[TMP3]]
-; CHECK-NEXT:    [[RET:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> [[C]])
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ne <16 x i16> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ne <16 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp ne <16 x i16> [[B]], zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp ne <16 x i16> [[C]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = and <16 x i1> [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP9:%.*]] = and <16 x i1> [[TMP6]], [[TMP5]]
+; CHECK-NEXT:    [[TMP10:%.*]] = and <16 x i1> [[TMP4]], [[TMP7]]
+; CHECK-NEXT:    [[TMP11:%.*]] = or <16 x i1> [[TMP8]], [[TMP9]]
+; CHECK-NEXT:    [[TMP12:%.*]] = or <16 x i1> [[TMP11]], [[TMP10]]
+; CHECK-NEXT:    [[TMP13:%.*]] = sext <16 x i1> [[TMP12]] to <16 x i16>
+; CHECK-NEXT:    [[TMP14:%.*]] = bitcast <16 x i16> [[TMP13]] to <8 x i32>
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp ne <8 x i32> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[TMP16:%.*]] = sext <8 x i1> [[TMP15]] to <8 x i32>
+; CHECK-NEXT:    [[_MSPROP1:%.*]] = or <8 x i32> [[TMP16]], [[TMP1]]
+; CHECK-NEXT:    [[RET:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> [[A]], <16 x i16> [[B]], <16 x i16> [[C]])
 ; CHECK-NEXT:    store <8 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8
 ; CHECK-NEXT:    ret <8 x i32> [[RET]]
 ;
-  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+  %ret = call <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)
   ret <8 x i32> %ret
 }
-declare <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> %A, <8 x i32> %B, <8 x i32> %C)
+declare <8 x i32> @llvm.x86.avx2.vpdpwuuds.256(<8 x i32> %A, <16 x i16> %B, <16 x i16> %C)

>From fc78c90ab2bc408efb29e5fad31e67a136b3984b Mon Sep 17 00:00:00 2001
From: BaiXilin <x53bai at uwaterloo.ca>
Date: Wed, 3 Dec 2025 23:48:05 -0500
Subject: [PATCH 3/4] header file comment format

---
 clang/lib/Headers/avxvnniint16intrin.h | 168 ++++++++++++-------------
 1 file changed, 81 insertions(+), 87 deletions(-)

diff --git a/clang/lib/Headers/avxvnniint16intrin.h b/clang/lib/Headers/avxvnniint16intrin.h
index 801795ac543dd..98d94ee3fcf3a 100644
--- a/clang/lib/Headers/avxvnniint16intrin.h
+++ b/clang/lib/Headers/avxvnniint16intrin.h
@@ -15,11 +15,11 @@
 #ifndef __AVXVNNIINT16INTRIN_H
 #define __AVXVNNIINT16INTRIN_H
 
-// clang-format off
 /// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a __A with
-///    corresponding unsigned 16-bit integers in \a __B, producing 2 intermediate
-///    signed 16-bit results. Sum these 2 results with the corresponding
-///    32-bit integer in \a __W, and store the packed 32-bit results in \a dst.
+///    corresponding unsigned 16-bit integers in \a __B, producing 2
+///    intermediate signed 16-bit results. Sum these 2 results with the
+///    corresponding 32-bit integer in \a __W, and store the packed 32-bit
+///    results in \a dst.
 ///
 /// \headerfile <immintrin.h>
 ///
@@ -41,21 +41,21 @@
 /// \code{.operation}
 /// FOR j := 0 to 3
 /// 	tmp1.dword := SignExtend32(__A.word[2*j]) * ZeroExtend32(__B.word[2*j])
-/// 	tmp2.dword := SignExtend32(__A.word[2*j+1]) * ZeroExtend32(__B.word[2*j+1])
+/// 	tmp2.dword :=
+///		  SignExtend32(__A.word[2*j+1]) * ZeroExtend32(__B.word[2*j+1])
 /// 	dst.dword[j] := __W.dword[j] + tmp1 + tmp2
 /// ENDFOR
 /// dst[MAX:128] := 0
 /// \endcode
-// clang-format on
 #define _mm_dpwsud_epi32(__W, __A, __B)                                        \
   ((__m128i)__builtin_ia32_vpdpwsud128((__v4si)(__W), (__v8hi)(__A),           \
                                        (__v8hu)(__B)))
 
-// clang-format off
 /// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a __A with
-///    corresponding unsigned 16-bit integers in \a __B, producing 2 intermediate
-///    signed 16-bit results. Sum these 2 results with the corresponding
-///    32-bit integer in \a __W, and store the packed 32-bit results in \a dst.
+///    corresponding unsigned 16-bit integers in \a __B, producing 2
+///    intermediate signed 16-bit results. Sum these 2 results with the
+///    corresponding 32-bit integer in \a __W, and store the packed 32-bit
+///    results in \a dst.
 ///
 /// \headerfile <immintrin.h>
 ///
@@ -77,22 +77,21 @@
 /// \code{.operation}
 /// FOR j := 0 to 7
 /// 	tmp1.dword := SignExtend32(__A.word[2*j]) * ZeroExtend32(__B.word[2*j])
-/// 	tmp2.dword := SignExtend32(__A.word[2*j+1]) * ZeroExtend32(__B.word[2*j+1])
-/// 	dst.dword[j] := __W.dword[j] + tmp1 + tmp2
+/// 	tmp2.dword :=
+///		  SignExtend32(__A.word[2*j+1]) * ZeroExtend32(__B.word[2*j+1])
+///		dst.dword[j] := __W.dword[j] + tmp1 + tmp2
 /// ENDFOR
 /// dst[MAX:256] := 0
 /// \endcode
-// clang-format on
 #define _mm256_dpwsud_epi32(__W, __A, __B)                                     \
   ((__m256i)__builtin_ia32_vpdpwsud256((__v8si)(__W), (__v16hi)(__A),          \
                                        (__v16hu)(__B)))
 
-// clang-format off
 /// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a __A with
-///    corresponding unsigned 16-bit integers in \a __B, producing 2 intermediate
-///    signed 16-bit results. Sum these 2 results with the corresponding
-///    32-bit integer in \a __W with signed saturation, and store the packed
-///    32-bit results in \a dst.
+///    corresponding unsigned 16-bit integers in \a __B, producing 2
+///    intermediate signed 16-bit results. Sum these 2 results with the
+///    corresponding 32-bit integer in \a __W with signed saturation, and store
+///    the packed 32-bit results in \a dst.
 ///
 /// \headerfile <immintrin.h>
 ///
@@ -114,23 +113,22 @@
 /// \code{.operation}
 /// FOR j := 0 to 3
 /// 	tmp1.dword := SignExtend32(__A.word[2*j]) * ZeroExtend32(__B.word[2*j])
-/// 	tmp2.dword := SignExtend32(__A.word[2*j+1]) * ZeroExtend32(__B.word[2*j+1])
+/// 	tmp2.dword :=
+///		  SignExtend32(__A.word[2*j+1]) * ZeroExtend32(__B.word[2*j+1])
 /// 	dst.dword[j] := SIGNED_DWORD_SATURATE(__W.dword[j] + tmp1 + tmp2)
 /// ENDFOR
 /// dst[MAX:128] := 0
 /// \endcode
-// clang-format on
 /// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a __A with
 #define _mm_dpwsuds_epi32(__W, __A, __B)                                       \
   ((__m128i)__builtin_ia32_vpdpwsuds128((__v4si)(__W), (__v8hi)(__A),          \
                                         (__v8hu)(__B)))
 
-// clang-format off
 /// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a __A with
-///    corresponding unsigned 16-bit integers in \a __B, producing 2 intermediate
-///    signed 16-bit results. Sum these 2 results with the corresponding
-///    32-bit integer in \a __W with signed saturation, and store the packed
-///    32-bit results in \a dst.
+///    corresponding unsigned 16-bit integers in \a __B, producing 2
+///    intermediate signed 16-bit results. Sum these 2 results with the
+///    corresponding 32-bit integer in \a __W with signed saturation, and store
+///    the packed 32-bit results in \a dst.
 ///
 /// \headerfile <immintrin.h>
 ///
@@ -152,21 +150,21 @@
 /// \code{.operation}
 /// FOR j := 0 to 7
 /// 	tmp1.dword := SignExtend32(__A.word[2*j]) * ZeroExtend32(__B.word[2*j])
-/// 	tmp2.dword := SignExtend32(__A.word[2*j+1]) * ZeroExtend32(__B.word[2*j+1])
+/// 	tmp2.dword :=
+///		  SignExtend32(__A.word[2*j+1]) * ZeroExtend32(__B.word[2*j+1])
 /// 	dst.dword[j] := SIGNED_DWORD_SATURATE(__W.dword[j] + tmp1 + tmp2)
 /// ENDFOR
 /// dst[MAX:256] := 0
 /// \endcode
-// clang-format on
 #define _mm256_dpwsuds_epi32(__W, __A, __B)                                    \
   ((__m256i)__builtin_ia32_vpdpwsuds256((__v8si)(__W), (__v16hi)(__A),         \
                                         (__v16hu)(__B)))
 
-// clang-format off
-/// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A with
-///    corresponding signed 16-bit integers in \a __B, producing 2 intermediate
-///    signed 16-bit results. Sum these 2 results with the corresponding
-///    32-bit integer in \a __W, and store the packed 32-bit results in \a dst.
+/// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A
+///    with corresponding signed 16-bit integers in \a __B, producing 2
+///    intermediate signed 16-bit results. Sum these 2 results with the
+///    corresponding 32-bit integer in \a __W, and store the packed 32-bit
+///    results in \a dst.
 ///
 /// \headerfile <immintrin.h>
 ///
@@ -188,21 +186,21 @@
 /// \code{.operation}
 /// FOR j := 0 to 3
 /// 	tmp1.dword := ZeroExtend32(__A.word[2*j]) * SignExtend32(__B.word[2*j])
-/// 	tmp2.dword := ZeroExtend32(__A.word[2*j+1]) * SignExtend32(__B.word[2*j+1])
+/// 	tmp2.dword :=
+/// 	  ZeroExtend32(__A.word[2*j+1]) * SignExtend32(__B.word[2*j+1])
 /// 	dst.dword[j] := __W.dword[j] + tmp1 + tmp2
 /// ENDFOR
 /// dst[MAX:128] := 0
 /// \endcode
-// clang-format on
 #define _mm_dpwusd_epi32(__W, __A, __B)                                        \
   ((__m128i)__builtin_ia32_vpdpwusd128((__v4si)(__W), (__v8hu)(__A),           \
                                        (__v8hi)(__B)))
 
-// clang-format off
-/// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A with
-///    corresponding signed 16-bit integers in \a __B, producing 2 intermediate
-///    signed 16-bit results. Sum these 2 results with the corresponding
-///    32-bit integer in \a __W, and store the packed 32-bit results in \a dst.
+/// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A
+///    with corresponding signed 16-bit integers in \a __B, producing 2
+///    intermediate signed 16-bit results. Sum these 2 results with the
+///    corresponding 32-bit integer in \a __W, and store the packed 32-bit
+///    results in \a dst.
 ///
 /// \headerfile <immintrin.h>
 ///
@@ -224,22 +222,21 @@
 /// \code{.operation}
 /// FOR j := 0 to 7
 /// 	tmp1.dword := ZeroExtend32(__A.word[2*j]) * SignExtend32(__B.word[2*j])
-/// 	tmp2.dword := ZeroExtend32(__A.word[2*j+1]) * SignExtend32(__B.word[2*j+1])
+/// 	tmp2.dword :=
+/// 	  ZeroExtend32(__A.word[2*j+1]) * SignExtend32(__B.word[2*j+1])
 /// 	dst.dword[j] := __W.dword[j] + tmp1 + tmp2
 /// ENDFOR
 /// dst[MAX:256] := 0
 /// \endcode
-// clang-format on
 #define _mm256_dpwusd_epi32(__W, __A, __B)                                     \
   ((__m256i)__builtin_ia32_vpdpwusd256((__v8si)(__W), (__v16hu)(__A),          \
                                        (__v16hi)(__B)))
 
-// clang-format off
-/// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A with
-///    corresponding signed 16-bit integers in \a __B, producing 2 intermediate
-///    signed 16-bit results. Sum these 2 results with the corresponding
-///    32-bit integer in \a __W with signed saturation, and store the packed
-///    32-bit results in \a dst.
+/// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A
+///    with corresponding signed 16-bit integers in \a __B, producing 2
+///    intermediate signed 16-bit results. Sum these 2 results with the
+///    corresponding 32-bit integer in \a __W with signed saturation, and
+///    store the packed 32-bit results in \a dst.
 ///
 /// \headerfile <immintrin.h>
 ///
@@ -261,22 +258,21 @@
 /// \code{.operation}
 /// FOR j := 0 to 3
 /// 	tmp1.dword := ZeroExtend32(__A.word[2*j]) * SignExtend32(__B.word[2*j])
-/// 	tmp2.dword := ZeroExtend32(__A.word[2*j+1]) * SignExtend32(__B.word[2*j+1])
+/// 	tmp2.dword :=
+/// 	  ZeroExtend32(__A.word[2*j+1]) * SignExtend32(__B.word[2*j+1])
 /// 	dst.dword[j] := SIGNED_DWORD_SATURATE(__W.dword[j] + tmp1 + tmp2)
 /// ENDFOR
 /// dst[MAX:128] := 0
 /// \endcode
-// clang-format on
 #define _mm_dpwusds_epi32(__W, __A, __B)                                       \
   ((__m128i)__builtin_ia32_vpdpwusds128((__v4si)(__W), (__v8hu)(__A),          \
                                         (__v8hi)(__B)))
 
-// clang-format off
-/// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A with
-///    corresponding signed 16-bit integers in \a __B, producing 2 intermediate
-///    signed 16-bit results. Sum these 2 results with the corresponding
-///    32-bit integer in \a __W with signed saturation, and store the packed
-///    32-bit results in \a dst.
+/// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A
+///    with corresponding signed 16-bit integers in \a __B, producing 2
+///    intermediate signed 16-bit results. Sum these 2 results with the
+///    corresponding 32-bit integer in \a __W with signed saturation, and
+///    store the packed 32-bit results in \a dst.
 ///
 /// \headerfile <immintrin.h>
 ///
@@ -298,21 +294,21 @@
 /// \code{.operation}
 /// FOR j := 0 to 7
 /// 	tmp1.dword := ZeroExtend32(__A.word[2*j]) * SignExtend32(__B.word[2*j])
-/// 	tmp2.dword := ZeroExtend32(__A.word[2*j+1]) * SignExtend32(__B.word[2*j+1])
+/// 	tmp2.dword :=
+/// 	  ZeroExtend32(__A.word[2*j+1]) * SignExtend32(__B.word[2*j+1])
 /// 	dst.dword[j] := SIGNED_DWORD_SATURATE(__W.dword[j] + tmp1 + tmp2)
 /// ENDFOR
 /// dst[MAX:256] := 0
 /// \endcode
-// clang-format on
 #define _mm256_dpwusds_epi32(__W, __A, __B)                                    \
   ((__m256i)__builtin_ia32_vpdpwusds256((__v8si)(__W), (__v16hu)(__A),         \
                                         (__v16hi)(__B)))
 
-// clang-format off
-/// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A with
-///    corresponding unsigned 16-bit integers in \a __B, producing 2 intermediate
-///    signed 16-bit results. Sum these 2 results with the corresponding
-///    32-bit integer in \a __W, and store the packed 32-bit results in \a dst.
+/// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A
+///    with corresponding unsigned 16-bit integers in \a __B, producing 2
+///    intermediate signed 16-bit results. Sum these 2 results with the
+///    corresponding 32-bit integer in \a __W, and store the packed 32-bit
+///    results in \a dst.
 ///
 /// \headerfile <immintrin.h>
 ///
@@ -334,21 +330,21 @@
 /// \code{.operation}
 /// FOR j := 0 to 3
 /// 	tmp1.dword := ZeroExtend32(__A.word[2*j]) * ZeroExtend32(__B.word[2*j])
-/// 	tmp2.dword := ZeroExtend32(__A.word[2*j+1]) * ZeroExtend32(__B.word[2*j+1])
+/// 	tmp2.dword :=
+/// 	  ZeroExtend32(__A.word[2*j+1]) * ZeroExtend32(__B.word[2*j+1])
 /// 	dst.dword[j] := __W.dword[j] + tmp1 + tmp2
 /// ENDFOR
 /// dst[MAX:128] := 0
 /// \endcode
-// clang-format on
 #define _mm_dpwuud_epi32(__W, __A, __B)                                        \
   ((__m128i)__builtin_ia32_vpdpwuud128((__v4si)(__W), (__v8hu)(__A),           \
                                        (__v8hu)(__B)))
 
-// clang-format off
-/// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A with
-///    corresponding unsigned 16-bit integers in \a __B, producing 2 intermediate
-///    signed 16-bit results. Sum these 2 results with the corresponding
-///    32-bit integer in \a __W, and store the packed 32-bit results in \a dst.
+/// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A
+///    with corresponding unsigned 16-bit integers in \a __B, producing 2
+///    intermediate signed 16-bit results. Sum these 2 results with the
+///    corresponding 32-bit integer in \a __W, and store the packed 32-bit
+///    results in \a dst.
 ///
 /// \headerfile <immintrin.h>
 ///
@@ -370,22 +366,21 @@
 /// \code{.operation}
 /// FOR j := 0 to 7
 /// 	tmp1.dword := ZeroExtend32(__A.word[2*j]) * ZeroExtend32(__B.word[2*j])
-/// 	tmp2.dword := ZeroExtend32(__A.word[2*j+1]) * ZeroExtend32(__B.word[2*j+1])
+/// 	tmp2.dword :=
+/// 	  ZeroExtend32(__A.word[2*j+1]) * ZeroExtend32(__B.word[2*j+1])
 /// 	dst.dword[j] := __W.dword[j] + tmp1 + tmp2
 /// ENDFOR
 /// dst[MAX:256] := 0
 /// \endcode
-// clang-format on
 #define _mm256_dpwuud_epi32(__W, __A, __B)                                     \
   ((__m256i)__builtin_ia32_vpdpwuud256((__v8si)(__W), (__v16hu)(__A),          \
                                        (__v16hu)(__B)))
 
-// clang-format off
-/// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A with
-///    corresponding unsigned 16-bit integers in \a __B, producing 2 intermediate
-///    signed 16-bit results. Sum these 2 results with the corresponding
-///    32-bit integer in \a __W with signed saturation, and store the packed
-///    32-bit results in \a dst.
+/// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A
+///    with corresponding unsigned 16-bit integers in \a __B, producing 2
+///    intermediate signed 16-bit results. Sum these 2 results with the
+///    corresponding 32-bit integer in \a __W with signed saturation, and store
+///    the packed 32-bit results in \a dst.
 ///
 /// \headerfile <immintrin.h>
 ///
@@ -407,22 +402,21 @@
 /// \code{.operation}
 /// FOR j := 0 to 3
 /// 	tmp1.dword := ZeroExtend32(__A.word[2*j]) * ZeroExtend32(__B.word[2*j])
-/// 	tmp2.dword := ZeroExtend32(__A.word[2*j+1]) * ZeroExtend32(__B.word[2*j+1])
+/// 	tmp2.dword :=
+/// 	  ZeroExtend32(__A.word[2*j+1]) * ZeroExtend32(__B.word[2*j+1])
 /// 	dst.dword[j] := UNSIGNED_DWORD_SATURATE(__W.dword[j] + tmp1 + tmp2)
 /// ENDFOR
 /// dst[MAX:128] := 0
 /// \endcode
-// clang-format on
 #define _mm_dpwuuds_epi32(__W, __A, __B)                                       \
   ((__m128i)__builtin_ia32_vpdpwuuds128((__v4si)(__W), (__v8hu)(__A),          \
                                         (__v8hu)(__B)))
 
-// clang-format off
-/// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A with
-///    corresponding unsigned 16-bit integers in \a __B, producing 2 intermediate
-///    signed 16-bit results. Sum these 2 results with the corresponding
-///    32-bit integer in \a __W with signed saturation, and store the packed
-///    32-bit results in \a dst.
+/// Multiply groups of 2 adjacent pairs of unsigned 16-bit integers in \a __A
+///    with corresponding unsigned 16-bit integers in \a __B, producing 2
+///    intermediate signed 16-bit results. Sum these 2 results with the
+///    corresponding 32-bit integer in \a __W with signed saturation, and store
+///    the packed 32-bit results in \a dst.
 ///
 /// \headerfile <immintrin.h>
 ///
@@ -444,12 +438,12 @@
 /// \code{.operation}
 /// FOR j := 0 to 7
 /// 	tmp1.dword := ZeroExtend32(__A.word[2*j]) * ZeroExtend32(__B.word[2*j])
-/// 	tmp2.dword := ZeroExtend32(__A.word[2*j+1]) * ZeroExtend32(__B.word[2*j+1])
+/// 	tmp2.dword :=
+/// 	  ZeroExtend32(__A.word[2*j+1]) * ZeroExtend32(__B.word[2*j+1])
 /// 	dst.dword[j] := UNSIGNED_DWORD_SATURATE(__W.dword[j] + tmp1 + tmp2)
 /// ENDFOR
 /// dst[MAX:256] := 0
 /// \endcode
-// clang-format on
 #define _mm256_dpwuuds_epi32(__W, __A, __B)                                    \
   ((__m256i)__builtin_ia32_vpdpwuuds256((__v8si)(__W), (__v16hu)(__A),         \
                                         (__v16hu)(__B)))

>From 03d70be63f41eba7cbb4284375a5aa4a14fad183 Mon Sep 17 00:00:00 2001
From: BaiXilin <x53bai at uwaterloo.ca>
Date: Mon, 8 Dec 2025 23:04:43 -0500
Subject: [PATCH 4/4] Generate masks to test mask.vpdpwssd[s].512

---
 .../X86/avx512vnni-intrinsics-upgrade.ll      | 38 +++++++++++++------
 1 file changed, 26 insertions(+), 12 deletions(-)

diff --git a/llvm/test/CodeGen/X86/avx512vnni-intrinsics-upgrade.ll b/llvm/test/CodeGen/X86/avx512vnni-intrinsics-upgrade.ll
index 35363e99ef4c8..2aabfab1c8666 100644
--- a/llvm/test/CodeGen/X86/avx512vnni-intrinsics-upgrade.ll
+++ b/llvm/test/CodeGen/X86/avx512vnni-intrinsics-upgrade.ll
@@ -116,12 +116,19 @@ define <16 x i32>@test_int_x86_avx512_vpdpwssd(<16 x i32> %x0, <16 x i32> %x1, <
 declare <16 x i32> @llvm.x86.avx512.mask.vpdpwssd.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
 declare <16 x i32> @llvm.x86.avx512.maskz.vpdpwssd.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
 
-define <16 x i32>@test_int_x86_avx512_mask_vpdpwssd_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
-; CHECK-LABEL: test_int_x86_avx512_mask_vpdpwssd_512:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vpdpwssd %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0x52,0xc2]
-; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <16 x i32> @llvm.x86.avx512.mask.vpdpwssd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 -1)
+define <16 x i32>@test_int_x86_avx512_mask_vpdpwssd_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x3) {
+; X86-LABEL: test_int_x86_avx512_mask_vpdpwssd_512:
+; X86:       # %bb.0:
+; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
+; X86-NEXT:    vpdpwssd %zmm2, %zmm1, %zmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x52,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_mask_vpdpwssd_512:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vpdpwssd %zmm2, %zmm1, %zmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x52,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <16 x i32> @llvm.x86.avx512.mask.vpdpwssd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x3)
   ret <16 x i32> %res
 }
 
@@ -166,12 +173,19 @@ define <16 x i32>@test_int_x86_avx512_vpdpwssds_512(<16 x i32> %x0, <16 x i32> %
 declare <16 x i32> @llvm.x86.avx512.mask.vpdpwssds.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
 declare <16 x i32> @llvm.x86.avx512.maskz.vpdpwssds.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
 
-define <16 x i32>@test_int_x86_avx512_mask_vpdpwssds_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
-; CHECK-LABEL: test_int_x86_avx512_mask_vpdpwssds_512:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vpdpwssds %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0x53,0xc2]
-; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <16 x i32> @llvm.x86.avx512.mask.vpdpwssds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 -1)
+define <16 x i32>@test_int_x86_avx512_mask_vpdpwssds_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x3) {
+; X86-LABEL: test_int_x86_avx512_mask_vpdpwssds_512:
+; X86:       # %bb.0:
+; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
+; X86-NEXT:    vpdpwssds %zmm2, %zmm1, %zmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x53,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_mask_vpdpwssds_512:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vpdpwssds %zmm2, %zmm1, %zmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x49,0x53,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <16 x i32> @llvm.x86.avx512.mask.vpdpwssds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x3)
   ret <16 x i32> %res
 }
 



More information about the llvm-commits mailing list