[clang] [clang][x86] Add constexpr support for VNNI intrinsics (PR #190549)

[Simon Pilgrim via cfe-commits]

================
@@ -3,165 +3,463 @@
 // RUN: %clang_cc1 -x c++ -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +avx512vnni -target-feature +avx512vl -emit-llvm -o - -Wall -Werror | FileCheck %s
 // RUN: %clang_cc1 -x c++ -flax-vector-conversions=none -ffreestanding %s -triple=i386-apple-darwin -target-feature +avx512vnni -target-feature +avx512vl -emit-llvm -o - -Wall -Werror | FileCheck %s
 
+// RUN: %clang_cc1 -x c -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +avx512vnni -target-feature +avx512vl -emit-llvm -o - -Wall -Werror -fexperimental-new-constant-interpreter | FileCheck %s
+// RUN: %clang_cc1 -x c++ -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +avx512vnni -target-feature +avx512vl -emit-llvm -o - -Wall -Werror -fexperimental-new-constant-interpreter | FileCheck %s
+// RUN: %clang_cc1 -x c -flax-vector-conversions=none -ffreestanding %s -triple=i386-apple-darwin -target-feature +avx512vnni -target-feature +avx512vl -emit-llvm -o - -Wall -Werror -fexperimental-new-constant-interpreter | FileCheck %s
+// RUN: %clang_cc1 -x c++ -flax-vector-conversions=none -ffreestanding %s -triple=i386-apple-darwin -target-feature +avx512vnni -target-feature +avx512vl -emit-llvm -o - -Wall -Werror -fexperimental-new-constant-interpreter | FileCheck %s
+
 #include <immintrin.h>
+#include "builtin_test_helpers.h"
 
 __m256i test_mm256_mask_dpbusd_epi32(__m256i __S, __mmask8 __U, __m256i __A, __m256i __B) {
   // CHECK-LABEL: test_mm256_mask_dpbusd_epi32
   // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32> %{{.*}}, <32 x i8> %{{.*}}, <32 x i8> %{{.*}})
   // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_mask_dpbusd_epi32(__S, __U, __A, __B);
 }
+TEST_CONSTEXPR(match_v8si(
+  _mm256_mask_dpbusd_epi32(
+    (__m256i)(__v8si){100, 200, 300, 400, 500, 600, 700, 800},
+    (__mmask8)0x55,
+    (__m256i)(__v32qu){1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1},
+    (__m256i)(__v32qi){1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}),
+  104, 200, 304, 400, 504, 600, 704, 800));
 
 __m256i test_mm256_maskz_dpbusd_epi32(__mmask8 __U, __m256i __S, __m256i __A, __m256i __B) {
   // CHECK-LABEL: test_mm256_maskz_dpbusd_epi32
   // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32> %{{.*}}, <32 x i8> %{{.*}}, <32 x i8> %{{.*}})
   // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
   return _mm256_maskz_dpbusd_epi32(__U, __S, __A, __B);
 }
+TEST_CONSTEXPR(match_v8si(
+  _mm256_maskz_dpbusd_epi32(
+    (__mmask8)0x0F,
+    (__m256i)(__v8si){0, 0, 0, 0, 0, 0, 0, 0},
+    (__m256i)(__v32qu){1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1},
+    (__m256i)(__v32qi){1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}),
+  4, 4, 4, 4, 0, 0, 0, 0));
 
 __m256i test_mm256_dpbusd_epi32(__m256i __S, __m256i __A, __m256i __B) {
   // CHECK-LABEL: test_mm256_dpbusd_epi32
   // CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32> %{{.*}}, <32 x i8> %{{.*}}, <32 x i8> %{{.*}})
   return _mm256_dpbusd_epi32(__S, __A, __B);
 }
+TEST_CONSTEXPR(match_v8si(
+  _mm256_dpbusd_epi32(
+    ((__m256i)(__v8si){0, 0, 0, 0, 0, 0, 0, 0}),
+    ((__m256i)(__v32qu){1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}),
+    ((__m256i)(__v32qi){1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1})),
+  4, 4, 4, 4, 4, 4, 4, 4));
+TEST_CONSTEXPR(match_v8si(
+  _mm256_dpbusd_epi32(
+    ((__m256i)(__v8si){10, 10, 10, 10, 10, 10, 10, 10}),
+    ((__m256i)(__v32qu){1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}),
+    ((__m256i)(__v32qi){1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1})),
+  14, 14, 14, 14, 14, 14, 14, 14));
+TEST_CONSTEXPR(match_v8si(
+  _mm256_dpbusd_epi32(
+    ((__m256i)(__v8si){0, 0, 0, 0, 0, 0, 0, 0}),
+    ((__m256i)(__v32qu){255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255}),
+    ((__m256i)(__v32qi){-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1})),
+  -1020, -1020, -1020, -1020, -1020, -1020, -1020, -1020));
+TEST_CONSTEXPR(match_v8si(
+  _mm256_dpbusd_epi32(
+    ((__m256i)(__v8si){2147483647, 2147483647, 2147483647, 2147483647, 2147483647, 2147483647, 2147483647, 2147483647}),
+    ((__m256i)(__v32qu){1,0,0,0,1,0,0,0,1,0,0,0,1,0,0,0,1,0,0,0,1,0,0,0,1,0,0,0,1,0,0,0}),
+    ((__m256i)(__v32qi){1,0,0,0,1,0,0,0,1,0,0,0,1,0,0,0,1,0,0,0,1,0,0,0,1,0,0,0,1,0,0,0})),
+  -2147483647-1, -2147483647-1, -2147483647-1, -2147483647-1, -2147483647-1, -2147483647-1, -2147483647-1, -2147483647-1));
----------------
RKSimon wrote:

We've generally tried to avoid uniform values in tests as it means we can miss cross-element mismatches in the evaluations - it also means we can reduce the number of TEST_CONSTEXPR entries

https://github.com/llvm/llvm-project/pull/190549