[llvm] [X86] Fold UNPCKL sign extension idioms to SIGN_EXTEND_VECTOR_INREG (PR #211979)
via llvm-commits
llvm-commits at lists.llvm.org
Fri Jul 24 20:00:34 PDT 2026
https://github.com/AZero13 created https://github.com/llvm/llvm-project/pull/211979
This patch adds a target-specific DAG combine to recognize common vector sign extension patterns that appear as UNPCKL shuffles and folds them directly into ISD::SIGN_EXTEND_VECTOR_INREG.
Specifically, it looks for an UNPCKL operation where the upper half is a sign mask of the lower half.
>From 19ab14a6c716f9f1244448d9c0180ea231182589 Mon Sep 17 00:00:00 2001
From: AZero13 <gfunni234 at gmail.com>
Date: Fri, 24 Jul 2026 18:19:31 -0400
Subject: [PATCH 1/2] [X86] Precommit tests for PMOVSX from unpack patterns
---
llvm/test/CodeGen/X86/pmovsx-unpack.ll | 88 ++++++++++++++++++++++++++
1 file changed, 88 insertions(+)
create mode 100644 llvm/test/CodeGen/X86/pmovsx-unpack.ll
diff --git a/llvm/test/CodeGen/X86/pmovsx-unpack.ll b/llvm/test/CodeGen/X86/pmovsx-unpack.ll
new file mode 100644
index 0000000000000..b1275ad16bb87
--- /dev/null
+++ b/llvm/test/CodeGen/X86/pmovsx-unpack.ll
@@ -0,0 +1,88 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-apple-macos -mattr=+sse2 | FileCheck %s --check-prefixes=SSE2
+; RUN: llc < %s -mtriple=x86_64-apple-macos -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE41
+; RUN: llc < %s -mtriple=x86_64-apple-macos -mattr=+avx2 | FileCheck %s --check-prefixes=AVX2
+
+define <8 x i16> @pmovsxbw_test(<16 x i8> %arg) {
+; SSE2-LABEL: pmovsxbw_test:
+; SSE2: ## %bb.0:
+; SSE2-NEXT: pxor %xmm1, %xmm1
+; SSE2-NEXT: pcmpgtb %xmm0, %xmm1
+; SSE2-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
+; SSE2-NEXT: retq
+;
+; SSE41-LABEL: pmovsxbw_test:
+; SSE41: ## %bb.0:
+; SSE41-NEXT: pxor %xmm1, %xmm1
+; SSE41-NEXT: pcmpgtb %xmm0, %xmm1
+; SSE41-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
+; SSE41-NEXT: retq
+;
+; AVX2-LABEL: pmovsxbw_test:
+; AVX2: ## %bb.0:
+; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
+; AVX2-NEXT: vpcmpgtb %xmm0, %xmm1, %xmm1
+; AVX2-NEXT: vpunpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
+; AVX2-NEXT: retq
+ %cmp = icmp sgt <16 x i8> zeroinitializer, %arg
+ %ext = sext <16 x i1> %cmp to <16 x i8>
+ %s = shufflevector <16 x i8> %arg, <16 x i8> %ext, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23>
+ %b = bitcast <16 x i8> %s to <8 x i16>
+ ret <8 x i16> %b
+}
+
+define <4 x i32> @pmovsxwd_test(<8 x i16> %arg) {
+; SSE2-LABEL: pmovsxwd_test:
+; SSE2: ## %bb.0:
+; SSE2-NEXT: pxor %xmm1, %xmm1
+; SSE2-NEXT: pcmpgtw %xmm0, %xmm1
+; SSE2-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; SSE2-NEXT: retq
+;
+; SSE41-LABEL: pmovsxwd_test:
+; SSE41: ## %bb.0:
+; SSE41-NEXT: pxor %xmm1, %xmm1
+; SSE41-NEXT: pcmpgtw %xmm0, %xmm1
+; SSE41-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; SSE41-NEXT: retq
+;
+; AVX2-LABEL: pmovsxwd_test:
+; AVX2: ## %bb.0:
+; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
+; AVX2-NEXT: vpcmpgtw %xmm0, %xmm1, %xmm1
+; AVX2-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX2-NEXT: retq
+ %cmp = icmp sgt <8 x i16> zeroinitializer, %arg
+ %ext = sext <8 x i1> %cmp to <8 x i16>
+ %s = shufflevector <8 x i16> %arg, <8 x i16> %ext, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11>
+ %b = bitcast <8 x i16> %s to <4 x i32>
+ ret <4 x i32> %b
+}
+
+define <2 x i64> @pmovsxdq_test(<4 x i32> %arg) {
+; SSE2-LABEL: pmovsxdq_test:
+; SSE2: ## %bb.0:
+; SSE2-NEXT: pxor %xmm1, %xmm1
+; SSE2-NEXT: pcmpgtd %xmm0, %xmm1
+; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE2-NEXT: retq
+;
+; SSE41-LABEL: pmovsxdq_test:
+; SSE41: ## %bb.0:
+; SSE41-NEXT: pxor %xmm1, %xmm1
+; SSE41-NEXT: pcmpgtd %xmm0, %xmm1
+; SSE41-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE41-NEXT: retq
+;
+; AVX2-LABEL: pmovsxdq_test:
+; AVX2: ## %bb.0:
+; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
+; AVX2-NEXT: vpcmpgtd %xmm0, %xmm1, %xmm1
+; AVX2-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX2-NEXT: retq
+ %cmp = icmp sgt <4 x i32> zeroinitializer, %arg
+ %ext = sext <4 x i1> %cmp to <4 x i32>
+ %s = shufflevector <4 x i32> %arg, <4 x i32> %ext, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
+ %b = bitcast <4 x i32> %s to <2 x i64>
+ ret <2 x i64> %b
+}
>From dcf9335f52213122d7d99762ed45dd0845e060e6 Mon Sep 17 00:00:00 2001
From: AZero13 <gfunni234 at gmail.com>
Date: Fri, 24 Jul 2026 19:02:12 -0400
Subject: [PATCH 2/2] [X86] Fold UNPCKL sign extension idioms to
SIGN_EXTEND_VECTOR_INREG
This patch adds a target-specific DAG combine to recognize common vector sign extension patterns that appear as UNPCKL shuffles and folds them directly into ISD::SIGN_EXTEND_VECTOR_INREG.
Specifically, it looks for an UNPCKL operation where the upper half is a sign mask of the lower half.
---
llvm/lib/Target/X86/X86ISelLowering.cpp | 38 +++++++++++++
llvm/test/CodeGen/X86/sign-extend-unpckl.ll | 58 ++++++++++++++++++++
llvm/test/CodeGen/X86/vector-trunc-math.ll | 60 ++++++++++-----------
3 files changed, 126 insertions(+), 30 deletions(-)
create mode 100644 llvm/test/CodeGen/X86/sign-extend-unpckl.ll
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 620f92062ef30..806cd8f013f64 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -42965,6 +42965,44 @@ static SDValue combineTargetShuffle(SDValue N, const SDLoc &DL,
return SDValue();
}
+ case X86ISD::UNPCKL: {
+ // Fold UNPCKL(X, PCMPGT(0, X)) or UNPCKL(X, SRA(X, EltSize-1))
+ // -> SIGN_EXTEND_VECTOR_INREG(X)
+ SDValue N0 = peekThroughBitcasts(N->getOperand(0));
+ SDValue N1 = peekThroughBitcasts(N->getOperand(1));
+
+ if (N0.getValueType() != N->getValueType(0))
+ return SDValue();
+
+ bool IsSignMask = false;
+ if (N1.getOpcode() == X86ISD::PCMPGT &&
+ peekThroughBitcasts(N1.getOperand(1)) == N0 &&
+ ISD::isBuildVectorAllZeros(N1.getOperand(0).getNode()))
+ IsSignMask = true;
+ else if ((N1.getOpcode() == ISD::SRA || N1.getOpcode() == X86ISD::VSRAI) &&
+ peekThroughBitcasts(N1.getOperand(0)) == N0) {
+ if (auto *C = isConstOrConstSplat(N1.getOperand(1), /*AllowUndefs*/ true)) {
+ if (C->getAPIntValue() == (N0.getScalarValueSizeInBits() - 1))
+ IsSignMask = true;
+ }
+ }
+
+ if (IsSignMask) {
+ EVT VT = N->getValueType(0);
+ EVT SVT = VT.getVectorElementType();
+ if (VT.is128BitVector() && SVT.getSizeInBits() <= 32) {
+ EVT ExtSVT =
+ EVT::getIntegerVT(*DAG.getContext(), SVT.getSizeInBits() * 2);
+ EVT ExtVT = EVT::getVectorVT(*DAG.getContext(), ExtSVT,
+ VT.getVectorNumElements() / 2);
+ SDValue Ext = DAG.getNode(ISD::SIGN_EXTEND_VECTOR_INREG, DL, ExtVT,
+ DAG.getBitcast(VT, N0));
+ return DAG.getBitcast(VT, Ext);
+ }
+ }
+
+ return SDValue();
+ }
case X86ISD::VBROADCAST: {
SDValue Src = N.getOperand(0);
SDValue BC = peekThroughBitcasts(Src);
diff --git a/llvm/test/CodeGen/X86/sign-extend-unpckl.ll b/llvm/test/CodeGen/X86/sign-extend-unpckl.ll
new file mode 100644
index 0000000000000..5efd0654b76af
--- /dev/null
+++ b/llvm/test/CodeGen/X86/sign-extend-unpckl.ll
@@ -0,0 +1,58 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+avx2 | FileCheck %s --check-prefixes=AVX2
+
+; Standard PCMPGT(0, X) mask
+define <4 x i32> @test_pcmpgt_v8i16(<8 x i16> %x) {
+; SSE-LABEL: test_pcmpgt_v8i16:
+; SSE: # %bb.0:
+; SSE-NEXT: pmovsxwd %xmm0, %xmm0
+; SSE-NEXT: retq
+;
+; AVX2-LABEL: test_pcmpgt_v8i16:
+; AVX2: # %bb.0:
+; AVX2-NEXT: vpmovsxwd %xmm0, %xmm0
+; AVX2-NEXT: retq
+ %cmp = icmp sgt <8 x i16> zeroinitializer, %x
+ %ext = sext <8 x i1> %cmp to <8 x i16>
+ %unpack = shufflevector <8 x i16> %x, <8 x i16> %ext, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11>
+ %bc = bitcast <8 x i16> %unpack to <4 x i32>
+ ret <4 x i32> %bc
+}
+
+; SRA mask
+define <4 x i32> @test_sra_v8i16(<8 x i16> %x) {
+; SSE-LABEL: test_sra_v8i16:
+; SSE: # %bb.0:
+; SSE-NEXT: pmovsxwd %xmm0, %xmm0
+; SSE-NEXT: retq
+;
+; AVX2-LABEL: test_sra_v8i16:
+; AVX2: # %bb.0:
+; AVX2-NEXT: vpmovsxwd %xmm0, %xmm0
+; AVX2-NEXT: retq
+ %mask = ashr <8 x i16> %x, <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
+ %unpack = shufflevector <8 x i16> %x, <8 x i16> %mask, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11>
+ %bc = bitcast <8 x i16> %unpack to <4 x i32>
+ ret <4 x i32> %bc
+}
+
+; Bitcasted zero vector for PCMPGT
+define <4 x i32> @test_pcmpgt_bitcast_zero_v8i16(<8 x i16> %x) {
+ ; Using a bitcast to attempt to hide the zero vector from simple checks
+; SSE-LABEL: test_pcmpgt_bitcast_zero_v8i16:
+; SSE: # %bb.0:
+; SSE-NEXT: pmovsxwd %xmm0, %xmm0
+; SSE-NEXT: retq
+;
+; AVX2-LABEL: test_pcmpgt_bitcast_zero_v8i16:
+; AVX2: # %bb.0:
+; AVX2-NEXT: vpmovsxwd %xmm0, %xmm0
+; AVX2-NEXT: retq
+ %zero = bitcast <4 x i32> zeroinitializer to <8 x i16>
+ %cmp = icmp sgt <8 x i16> %zero, %x
+ %ext = sext <8 x i1> %cmp to <8 x i16>
+ %unpack = shufflevector <8 x i16> %x, <8 x i16> %ext, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11>
+ %bc = bitcast <8 x i16> %unpack to <4 x i32>
+ ret <4 x i32> %bc
+}
diff --git a/llvm/test/CodeGen/X86/vector-trunc-math.ll b/llvm/test/CodeGen/X86/vector-trunc-math.ll
index 2f07166df813a..950942724b756 100644
--- a/llvm/test/CodeGen/X86/vector-trunc-math.ll
+++ b/llvm/test/CodeGen/X86/vector-trunc-math.ll
@@ -458,14 +458,14 @@ define <4 x i32> @trunc_add_const_v4i64_v4i32(<4 x i64> %a0) nounwind {
; SSE-LABEL: trunc_add_const_v4i64_v4i32:
; SSE: # %bb.0:
; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
-; SSE-NEXT: paddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE-NEXT: paddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 # [0,1,2,3]
; SSE-NEXT: retq
;
; AVX1-LABEL: trunc_add_const_v4i64_v4i32:
; AVX1: # %bb.0:
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
; AVX1-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
-; AVX1-NEXT: vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT: vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [0,1,2,3]
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
;
@@ -473,7 +473,7 @@ define <4 x i32> @trunc_add_const_v4i64_v4i32(<4 x i64> %a0) nounwind {
; AVX2-SLOW: # %bb.0:
; AVX2-SLOW-NEXT: vextractf128 $1, %ymm0, %xmm1
; AVX2-SLOW-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
-; AVX2-SLOW-NEXT: vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-SLOW-NEXT: vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [0,1,2,3]
; AVX2-SLOW-NEXT: vzeroupper
; AVX2-SLOW-NEXT: retq
;
@@ -481,7 +481,7 @@ define <4 x i32> @trunc_add_const_v4i64_v4i32(<4 x i64> %a0) nounwind {
; AVX2-FAST-ALL: # %bb.0:
; AVX2-FAST-ALL-NEXT: vpmovsxbd {{.*#+}} ymm1 = [0,2,4,6,0,0,0,0]
; AVX2-FAST-ALL-NEXT: vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-ALL-NEXT: vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-FAST-ALL-NEXT: vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [0,1,2,3]
; AVX2-FAST-ALL-NEXT: vzeroupper
; AVX2-FAST-ALL-NEXT: retq
;
@@ -489,7 +489,7 @@ define <4 x i32> @trunc_add_const_v4i64_v4i32(<4 x i64> %a0) nounwind {
; AVX2-FAST-PERLANE: # %bb.0:
; AVX2-FAST-PERLANE-NEXT: vextractf128 $1, %ymm0, %xmm1
; AVX2-FAST-PERLANE-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
-; AVX2-FAST-PERLANE-NEXT: vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT: vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [0,1,2,3]
; AVX2-FAST-PERLANE-NEXT: vzeroupper
; AVX2-FAST-PERLANE-NEXT: retq
;
@@ -497,7 +497,7 @@ define <4 x i32> @trunc_add_const_v4i64_v4i32(<4 x i64> %a0) nounwind {
; AVX512: # %bb.0:
; AVX512-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
; AVX512-NEXT: vpmovqd %zmm0, %ymm0
-; AVX512-NEXT: vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX512-NEXT: vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [0,1,2,3]
; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq
%1 = add <4 x i64> %a0, <i64 0, i64 1, i64 2, i64 3>
@@ -515,7 +515,7 @@ define <8 x i16> @trunc_add_const_v8i64_v8i16(<8 x i64> %a0) nounwind {
; SSE-NEXT: pslld $16, %xmm0
; SSE-NEXT: psrad $16, %xmm0
; SSE-NEXT: packssdw %xmm2, %xmm0
-; SSE-NEXT: paddw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE-NEXT: paddw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 # [0,1,2,3,4,5,6,7]
; SSE-NEXT: retq
;
; AVX1-LABEL: trunc_add_const_v8i64_v8i16:
@@ -528,7 +528,7 @@ define <8 x i16> @trunc_add_const_v8i64_v8i16(<8 x i64> %a0) nounwind {
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2
; AVX1-NEXT: vpackusdw %xmm2, %xmm0, %xmm0
; AVX1-NEXT: vpackusdw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT: vpaddw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT: vpaddw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [0,1,2,3,4,5,6,7]
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
;
@@ -541,14 +541,14 @@ define <8 x i16> @trunc_add_const_v8i64_v8i16(<8 x i64> %a0) nounwind {
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX2-NEXT: vpackusdw %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
-; AVX2-NEXT: vpaddw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT: vpaddw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [0,1,2,3,4,5,6,7]
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
; AVX512-LABEL: trunc_add_const_v8i64_v8i16:
; AVX512: # %bb.0:
; AVX512-NEXT: vpmovqw %zmm0, %xmm0
-; AVX512-NEXT: vpaddw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX512-NEXT: vpaddw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [0,1,2,3,4,5,6,7]
; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq
%1 = add <8 x i64> %a0, <i64 0, i64 1, i64 2, i64 3, i64 4, i64 5, i64 6, i64 7>
@@ -564,7 +564,7 @@ define <8 x i16> @trunc_add_const_v8i32_v8i16(<8 x i32> %a0) nounwind {
; SSE-NEXT: pslld $16, %xmm0
; SSE-NEXT: psrad $16, %xmm0
; SSE-NEXT: packssdw %xmm1, %xmm0
-; SSE-NEXT: paddw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE-NEXT: paddw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 # [0,1,2,3,4,5,6,7]
; SSE-NEXT: retq
;
; AVX1-LABEL: trunc_add_const_v8i32_v8i16:
@@ -572,7 +572,7 @@ define <8 x i16> @trunc_add_const_v8i32_v8i16(<8 x i32> %a0) nounwind {
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
; AVX1-NEXT: vpackusdw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT: vpaddw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT: vpaddw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [0,1,2,3,4,5,6,7]
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
;
@@ -580,7 +580,7 @@ define <8 x i16> @trunc_add_const_v8i32_v8i16(<8 x i32> %a0) nounwind {
; AVX2: # %bb.0:
; AVX2-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u,16,17,20,21,24,25,28,29,u,u,u,u,u,u,u,u]
; AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT: vpaddw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT: vpaddw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [0,1,2,3,4,5,6,7]
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
@@ -588,7 +588,7 @@ define <8 x i16> @trunc_add_const_v8i32_v8i16(<8 x i32> %a0) nounwind {
; AVX512: # %bb.0:
; AVX512-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
; AVX512-NEXT: vpmovdw %zmm0, %ymm0
-; AVX512-NEXT: vpaddw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX512-NEXT: vpaddw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [0,1,2,3,4,5,6,7]
; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq
%1 = add <8 x i32> %a0, <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
@@ -615,7 +615,7 @@ define <16 x i8> @trunc_add_const_v16i64_v16i8(<16 x i64> %a0) nounwind {
; SSE-NEXT: packuswb %xmm1, %xmm0
; SSE-NEXT: packuswb %xmm2, %xmm0
; SSE-NEXT: packuswb %xmm4, %xmm0
-; SSE-NEXT: paddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE-NEXT: paddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 # [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
; SSE-NEXT: retq
;
; AVX1-LABEL: trunc_add_const_v16i64_v16i8:
@@ -636,7 +636,7 @@ define <16 x i8> @trunc_add_const_v16i64_v16i8(<16 x i64> %a0) nounwind {
; AVX1-NEXT: vpackusdw %xmm3, %xmm0, %xmm0
; AVX1-NEXT: vpackusdw %xmm1, %xmm0, %xmm0
; AVX1-NEXT: vpackuswb %xmm2, %xmm0, %xmm0
-; AVX1-NEXT: vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT: vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
;
@@ -655,7 +655,7 @@ define <16 x i8> @trunc_add_const_v16i64_v16i8(<16 x i64> %a0) nounwind {
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX2-NEXT: vpackuswb %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
-; AVX2-NEXT: vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT: vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
@@ -664,7 +664,7 @@ define <16 x i8> @trunc_add_const_v16i64_v16i8(<16 x i64> %a0) nounwind {
; AVX512-NEXT: vpmovqb %zmm1, %xmm1
; AVX512-NEXT: vpmovqb %zmm0, %xmm0
; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512-NEXT: vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX512-NEXT: vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq
%1 = add <16 x i64> %a0, <i64 0, i64 1, i64 2, i64 3, i64 4, i64 5, i64 6, i64 7, i64 8, i64 9, i64 10, i64 11, i64 12, i64 13, i64 14, i64 15>
@@ -683,7 +683,7 @@ define <16 x i8> @trunc_add_const_v16i32_v16i8(<16 x i32> %a0) nounwind {
; SSE-NEXT: pand %xmm4, %xmm0
; SSE-NEXT: packuswb %xmm1, %xmm0
; SSE-NEXT: packuswb %xmm2, %xmm0
-; SSE-NEXT: paddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE-NEXT: paddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 # [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
; SSE-NEXT: retq
;
; AVX1-LABEL: trunc_add_const_v16i32_v16i8:
@@ -696,7 +696,7 @@ define <16 x i8> @trunc_add_const_v16i32_v16i8(<16 x i32> %a0) nounwind {
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2
; AVX1-NEXT: vpackusdw %xmm2, %xmm0, %xmm0
; AVX1-NEXT: vpackuswb %xmm1, %xmm0, %xmm0
-; AVX1-NEXT: vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT: vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
;
@@ -709,14 +709,14 @@ define <16 x i8> @trunc_add_const_v16i32_v16i8(<16 x i32> %a0) nounwind {
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX2-NEXT: vpackuswb %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
-; AVX2-NEXT: vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT: vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
; AVX512-LABEL: trunc_add_const_v16i32_v16i8:
; AVX512: # %bb.0:
; AVX512-NEXT: vpmovdb %zmm0, %xmm0
-; AVX512-NEXT: vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX512-NEXT: vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq
%1 = add <16 x i32> %a0, <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
@@ -731,7 +731,7 @@ define <16 x i8> @trunc_add_const_v16i16_v16i8(<16 x i16> %a0) nounwind {
; SSE-NEXT: pand %xmm2, %xmm1
; SSE-NEXT: pand %xmm2, %xmm0
; SSE-NEXT: packuswb %xmm1, %xmm0
-; SSE-NEXT: paddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE-NEXT: paddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 # [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
; SSE-NEXT: retq
;
; AVX1-LABEL: trunc_add_const_v16i16_v16i8:
@@ -739,7 +739,7 @@ define <16 x i8> @trunc_add_const_v16i16_v16i8(<16 x i16> %a0) nounwind {
; AVX1-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
; AVX1-NEXT: vpackuswb %xmm1, %xmm0, %xmm0
-; AVX1-NEXT: vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT: vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
;
@@ -748,7 +748,7 @@ define <16 x i8> @trunc_add_const_v16i16_v16i8(<16 x i16> %a0) nounwind {
; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX2-NEXT: vpackuswb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT: vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT: vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
@@ -756,7 +756,7 @@ define <16 x i8> @trunc_add_const_v16i16_v16i8(<16 x i16> %a0) nounwind {
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
; AVX512F-NEXT: vpmovdb %zmm0, %xmm0
-; AVX512F-NEXT: vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX512F-NEXT: vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
;
@@ -764,7 +764,7 @@ define <16 x i8> @trunc_add_const_v16i16_v16i8(<16 x i16> %a0) nounwind {
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0
-; AVX512BW-NEXT: vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX512BW-NEXT: vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
; AVX512BW-NEXT: vzeroupper
; AVX512BW-NEXT: retq
;
@@ -772,7 +772,7 @@ define <16 x i8> @trunc_add_const_v16i16_v16i8(<16 x i16> %a0) nounwind {
; AVX512DQ: # %bb.0:
; AVX512DQ-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
; AVX512DQ-NEXT: vpmovdb %zmm0, %xmm0
-; AVX512DQ-NEXT: vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX512DQ-NEXT: vpaddb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15]
; AVX512DQ-NEXT: vzeroupper
; AVX512DQ-NEXT: retq
%1 = add <16 x i16> %a0, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>
@@ -4595,13 +4595,13 @@ define <4 x i32> @mul_add_const_v4i64_v4i32(<4 x i32> %a0, <4 x i32> %a1) nounwi
; SSE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,1,3,3]
; SSE-NEXT: pmuludq %xmm3, %xmm1
; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
-; SSE-NEXT: paddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE-NEXT: paddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 # [4294967293,4294967295,1,3]
; SSE-NEXT: retq
;
; AVX-LABEL: mul_add_const_v4i64_v4i32:
; AVX: # %bb.0:
; AVX-NEXT: vpmulld %xmm1, %xmm0, %xmm0
-; AVX-NEXT: vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX-NEXT: vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [4294967293,4294967295,1,3]
; AVX-NEXT: retq
%1 = sext <4 x i32> %a0 to <4 x i64>
%2 = sext <4 x i32> %a1 to <4 x i64>
More information about the llvm-commits
mailing list