[llvm] [X86] combine-pmaddwd.ll - test with x86-64 levels not raw SSE levels (PR #208666)

Simon Pilgrim via llvm-commits llvm-commits at lists.llvm.org
Fri Jul 10 02:06:50 PDT 2026


https://github.com/RKSimon created https://github.com/llvm/llvm-project/pull/208666

Add missing AVX512 coverage

>From 4925cae47eab0aa4cb7c6bea56be6c9c1e10a4a8 Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev at redking.me.uk>
Date: Fri, 10 Jul 2026 10:05:32 +0100
Subject: [PATCH] [X86] combine-pmaddwd.ll - test with x86-64 levels not raw
 SSE levels

Add missing AVX512 coverage
---
 llvm/test/CodeGen/X86/combine-pmaddwd.ll | 220 +++++++++++++++--------
 1 file changed, 142 insertions(+), 78 deletions(-)

diff --git a/llvm/test/CodeGen/X86/combine-pmaddwd.ll b/llvm/test/CodeGen/X86/combine-pmaddwd.ll
index d1e5e720af7fb..4fe3569df88e9 100644
--- a/llvm/test/CodeGen/X86/combine-pmaddwd.ll
+++ b/llvm/test/CodeGen/X86/combine-pmaddwd.ll
@@ -1,8 +1,9 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2   | FileCheck %s --check-prefixes=CHECK,SSE,SSE2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=CHECK,SSE,SSE41
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX,AVX1
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX,AVX2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64      | FileCheck %s --check-prefixes=CHECK,SSE,SSE2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v2   | FileCheck %s --check-prefixes=CHECK,SSE,SSE42
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=sandybridge | FileCheck %s --check-prefixes=CHECK,AVX,AVX1
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v3   | FileCheck %s --check-prefixes=CHECK,AVX,AVX2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v4   | FileCheck %s --check-prefixes=CHECK,AVX,AVX512
 
 define <4 x i32> @combine_pmaddwd_zero(<8 x i16> %a0, <8 x i16> %a1) {
 ; SSE-LABEL: combine_pmaddwd_zero:
@@ -53,6 +54,13 @@ define <8 x i32> @combine_pmaddwd_concat(<8 x i16> %a0, <8 x i16> %a1, <8 x i16>
 ; AVX2-NEXT:    vpmaddwd %xmm3, %xmm2, %xmm1
 ; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: combine_pmaddwd_concat:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpmaddwd %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpmaddwd %xmm3, %xmm2, %xmm1
+; AVX512-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
+; AVX512-NEXT:    retq
   %1 = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> %a0, <8 x i16> %a1)
   %2 = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> %a2, <8 x i16> %a3)
   %3 = shufflevector <4 x i32> %1, <4 x i32> %2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
@@ -60,19 +68,12 @@ define <8 x i32> @combine_pmaddwd_concat(<8 x i16> %a0, <8 x i16> %a1, <8 x i16>
 }
 
 define <8 x i32> @combine_pmaddwd_concat_freeze(<8 x i16> %a0, <8 x i16> %a1) {
-; SSE2-LABEL: combine_pmaddwd_concat_freeze:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [1,1,1,1,1,1,1,1]
-; SSE2-NEXT:    pmaddwd %xmm2, %xmm0
-; SSE2-NEXT:    pmaddwd %xmm2, %xmm1
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: combine_pmaddwd_concat_freeze:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    pmovsxbw {{.*#+}} xmm2 = [1,1,1,1,1,1,1,1]
-; SSE41-NEXT:    pmaddwd %xmm2, %xmm0
-; SSE41-NEXT:    pmaddwd %xmm2, %xmm1
-; SSE41-NEXT:    retq
+; SSE-LABEL: combine_pmaddwd_concat_freeze:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [1,1,1,1,1,1,1,1]
+; SSE-NEXT:    pmaddwd %xmm2, %xmm0
+; SSE-NEXT:    pmaddwd %xmm2, %xmm1
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: combine_pmaddwd_concat_freeze:
 ; AVX1:       # %bb.0:
@@ -88,6 +89,13 @@ define <8 x i32> @combine_pmaddwd_concat_freeze(<8 x i16> %a0, <8 x i16> %a1) {
 ; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpmaddwd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0 # [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
 ; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: combine_pmaddwd_concat_freeze:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
+; AVX512-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
+; AVX512-NEXT:    vpmaddwd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0 # [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
+; AVX512-NEXT:    retq
   %lo = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> %a0, <8 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>)
   %hi = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> %a1, <8 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>)
   %flo = freeze <4 x i32> %lo
@@ -113,11 +121,11 @@ define <8 x i32> @combine_pmaddwd_concat_add(ptr %px0, ptr %py0) {
 ; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
 ; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = mem[0,2,2,3,4,5,6,7]
 ; SSE2-NEXT:    pmaddwd %xmm1, %xmm0
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = mem[0,2,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm2[0,1,2,3,4,6,6,7]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = mem[0,2,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
 ; SSE2-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,6,6,7]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
 ; SSE2-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
@@ -131,32 +139,32 @@ define <8 x i32> @combine_pmaddwd_concat_add(ptr %px0, ptr %py0) {
 ; SSE2-NEXT:    pmaddwd %xmm2, %xmm1
 ; SSE2-NEXT:    retq
 ;
-; SSE41-LABEL: combine_pmaddwd_concat_add:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa (%rdi), %xmm0
-; SSE41-NEXT:    movdqa 16(%rdi), %xmm1
-; SSE41-NEXT:    movdqa 32(%rdi), %xmm2
-; SSE41-NEXT:    movdqa 48(%rdi), %xmm3
-; SSE41-NEXT:    movdqa (%rsi), %xmm4
-; SSE41-NEXT:    movdqa 16(%rsi), %xmm5
-; SSE41-NEXT:    movdqa 32(%rsi), %xmm6
-; SSE41-NEXT:    movdqa 48(%rsi), %xmm7
-; SSE41-NEXT:    movq {{.*#+}} xmm8 = [0,1,4,5,8,9,12,13,0,0,0,0,0,0,0,0]
-; SSE41-NEXT:    pshufb %xmm8, %xmm6
-; SSE41-NEXT:    pshufb %xmm8, %xmm4
-; SSE41-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm6[0],xmm4[1],xmm6[1],xmm4[2],xmm6[2],xmm4[3],xmm6[3]
-; SSE41-NEXT:    pshufb %xmm8, %xmm2
-; SSE41-NEXT:    pshufb %xmm8, %xmm0
-; SSE41-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; SSE41-NEXT:    pmaddwd %xmm4, %xmm0
-; SSE41-NEXT:    pshufb %xmm8, %xmm7
-; SSE41-NEXT:    pshufb %xmm8, %xmm5
-; SSE41-NEXT:    punpcklwd {{.*#+}} xmm5 = xmm5[0],xmm7[0],xmm5[1],xmm7[1],xmm5[2],xmm7[2],xmm5[3],xmm7[3]
-; SSE41-NEXT:    pshufb %xmm8, %xmm3
-; SSE41-NEXT:    pshufb %xmm8, %xmm1
-; SSE41-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3]
-; SSE41-NEXT:    pmaddwd %xmm5, %xmm1
-; SSE41-NEXT:    retq
+; SSE42-LABEL: combine_pmaddwd_concat_add:
+; SSE42:       # %bb.0:
+; SSE42-NEXT:    movdqa (%rdi), %xmm0
+; SSE42-NEXT:    movdqa 16(%rdi), %xmm1
+; SSE42-NEXT:    movdqa 32(%rdi), %xmm2
+; SSE42-NEXT:    movdqa 48(%rdi), %xmm3
+; SSE42-NEXT:    movdqa (%rsi), %xmm4
+; SSE42-NEXT:    movdqa 16(%rsi), %xmm5
+; SSE42-NEXT:    movdqa 32(%rsi), %xmm6
+; SSE42-NEXT:    movdqa 48(%rsi), %xmm7
+; SSE42-NEXT:    movq {{.*#+}} xmm8 = [0,1,4,5,8,9,12,13,0,0,0,0,0,0,0,0]
+; SSE42-NEXT:    pshufb %xmm8, %xmm6
+; SSE42-NEXT:    pshufb %xmm8, %xmm4
+; SSE42-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm6[0],xmm4[1],xmm6[1],xmm4[2],xmm6[2],xmm4[3],xmm6[3]
+; SSE42-NEXT:    pshufb %xmm8, %xmm2
+; SSE42-NEXT:    pshufb %xmm8, %xmm0
+; SSE42-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
+; SSE42-NEXT:    pmaddwd %xmm4, %xmm0
+; SSE42-NEXT:    pshufb %xmm8, %xmm7
+; SSE42-NEXT:    pshufb %xmm8, %xmm5
+; SSE42-NEXT:    punpcklwd {{.*#+}} xmm5 = xmm5[0],xmm7[0],xmm5[1],xmm7[1],xmm5[2],xmm7[2],xmm5[3],xmm7[3]
+; SSE42-NEXT:    pshufb %xmm8, %xmm3
+; SSE42-NEXT:    pshufb %xmm8, %xmm1
+; SSE42-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3]
+; SSE42-NEXT:    pmaddwd %xmm5, %xmm1
+; SSE42-NEXT:    retq
 ;
 ; AVX1-LABEL: combine_pmaddwd_concat_add:
 ; AVX1:       # %bb.0:
@@ -195,6 +203,16 @@ define <8 x i32> @combine_pmaddwd_concat_add(ptr %px0, ptr %py0) {
 ; AVX2-NEXT:    vpmaddwd 32(%rdi), %ymm0, %ymm0
 ; AVX2-NEXT:    vpaddd %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: combine_pmaddwd_concat_add:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpxor %xmm0, %xmm0, %xmm0
+; AVX512-NEXT:    vpblendw {{.*#+}} ymm1 = mem[0],ymm0[1],mem[2],ymm0[3],mem[4],ymm0[5],mem[6],ymm0[7],mem[8],ymm0[9],mem[10],ymm0[11],mem[12],ymm0[13],mem[14],ymm0[15]
+; AVX512-NEXT:    vpmaddwd (%rdi), %ymm1, %ymm1
+; AVX512-NEXT:    vpblendw {{.*#+}} ymm0 = mem[0],ymm0[1],mem[2],ymm0[3],mem[4],ymm0[5],mem[6],ymm0[7],mem[8],ymm0[9],mem[10],ymm0[11],mem[12],ymm0[13],mem[14],ymm0[15]
+; AVX512-NEXT:    vpmaddwd 32(%rdi), %ymm0, %ymm0
+; AVX512-NEXT:    vpaddd %ymm0, %ymm1, %ymm0
+; AVX512-NEXT:    retq
   %px1 = getelementptr inbounds nuw i8, ptr %px0, i64 16
   %py1 = getelementptr inbounds nuw i8, ptr %py0, i64 16
   %px2 = getelementptr inbounds nuw i8, ptr %px0, i64 32
@@ -241,6 +259,12 @@ define <4 x i32> @combine_pmaddwd_demandedelts(<8 x i16> %a0, <8 x i16> %a1) {
 ; AVX2-NEXT:    vpmaddwd %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpbroadcastd %xmm0, %xmm0
 ; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: combine_pmaddwd_demandedelts:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpmaddwd %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpbroadcastd %xmm0, %xmm0
+; AVX512-NEXT:    retq
   %1 = shufflevector <8 x i16> %a0, <8 x i16> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 4, i32 4, i32 4>
   %2 = shufflevector <8 x i16> %a1, <8 x i16> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 7, i32 7, i32 7, i32 7>
   %3 = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> %1, <8 x i16> %2)
@@ -283,10 +307,31 @@ define <4 x i32> @combine_pmaddwd_constant_nsw() {
 
 ; Constant folding PMADDWD was causing an infinite loop in the PCMPGT commuting between 2 constant values.
 define i1 @pmaddwd_pcmpgt_infinite_loop() {
-; CHECK-LABEL: pmaddwd_pcmpgt_infinite_loop:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    movb $1, %al
-; CHECK-NEXT:    retq
+; SSE-LABEL: pmaddwd_pcmpgt_infinite_loop:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movb $1, %al
+; SSE-NEXT:    retq
+;
+; AVX1-LABEL: pmaddwd_pcmpgt_infinite_loop:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    movb $1, %al
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: pmaddwd_pcmpgt_infinite_loop:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    movb $1, %al
+; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: pmaddwd_pcmpgt_infinite_loop:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpxor %xmm0, %xmm0, %xmm0
+; AVX512-NEXT:    vpbroadcastd {{.*#+}} xmm1 = [2147483647,2147483647,2147483647,2147483647]
+; AVX512-NEXT:    vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1 # [4294967288,4294967287,4294967286,4294967285]
+; AVX512-NEXT:    vpcmpgtd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm0, %k1
+; AVX512-NEXT:    vpcmpgtd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %k0 {%k1}
+; AVX512-NEXT:    kortestb %k0, %k0
+; AVX512-NEXT:    sete %al
+; AVX512-NEXT:    retq
   %1 = tail call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> <i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768>, <8 x i16> <i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768, i16 -32768>)
   %2 = icmp eq <4 x i32> %1, <i32 -2147483648, i32 -2147483648, i32 -2147483648, i32 -2147483648>
   %3 = select <4 x i1> %2, <4 x i32> <i32 2147483647, i32 2147483647, i32 2147483647, i32 2147483647>, <4 x i32> zeroinitializer
@@ -301,19 +346,12 @@ define i1 @pmaddwd_pcmpgt_infinite_loop() {
 
 ; If the shuffle matches, but there is no multiply, introduce a trivial multiply by 1.
 define <8 x i32> @sext_pairwise_add(<16 x i16> %x) {
-; SSE2-LABEL: sext_pairwise_add:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [1,1,1,1,1,1,1,1]
-; SSE2-NEXT:    pmaddwd %xmm2, %xmm0
-; SSE2-NEXT:    pmaddwd %xmm2, %xmm1
-; SSE2-NEXT:    retq
-;
-; SSE41-LABEL: sext_pairwise_add:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    pmovsxbw {{.*#+}} xmm2 = [1,1,1,1,1,1,1,1]
-; SSE41-NEXT:    pmaddwd %xmm2, %xmm0
-; SSE41-NEXT:    pmaddwd %xmm2, %xmm1
-; SSE41-NEXT:    retq
+; SSE-LABEL: sext_pairwise_add:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [1,1,1,1,1,1,1,1]
+; SSE-NEXT:    pmaddwd %xmm2, %xmm0
+; SSE-NEXT:    pmaddwd %xmm2, %xmm1
+; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: sext_pairwise_add:
 ; AVX1:       # %bb.0:
@@ -328,6 +366,11 @@ define <8 x i32> @sext_pairwise_add(<16 x i16> %x) {
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmaddwd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0 # [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
 ; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: sext_pairwise_add:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpmaddwd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0 # [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
+; AVX512-NEXT:    retq
   %1 = sext <16 x i16> %x to <16 x i32>
   %2 = shufflevector <16 x i32> %1, <16 x i32> poison, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
   %3 = shufflevector <16 x i32> %1, <16 x i32> poison, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
@@ -356,6 +399,11 @@ define <8 x i32> @combine_with_mul(<16 x i16> %v) {
 ; AVX2:       # %bb.0: # %bb1
 ; AVX2-NEXT:    vpmaddwd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0 # [4096,1,4096,1,4096,1,4096,1,4096,1,4096,1,4096,1,4096,1]
 ; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: combine_with_mul:
+; AVX512:       # %bb.0: # %bb1
+; AVX512-NEXT:    vpmaddwd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0 # [4096,1,4096,1,4096,1,4096,1,4096,1,4096,1,4096,1,4096,1]
+; AVX512-NEXT:    retq
 bb1:
   %0 = sext <16 x i16> %v to <16 x i32>
   %1 = mul nsw <16 x i32> %0, <i32 4096, i32 1, i32 4096, i32 1, i32 4096, i32 1, i32 4096, i32 1, i32 4096, i32 1, i32 4096, i32 1, i32 4096, i32 1, i32 4096, i32 1>
@@ -386,6 +434,11 @@ define <8 x i32> @combine_with_shl(<16 x i16> %v) {
 ; AVX2:       # %bb.0: # %bb1
 ; AVX2-NEXT:    vpmaddwd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0 # [4096,1,4096,1,4096,1,4096,1,4096,1,4096,1,4096,1,4096,1]
 ; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: combine_with_shl:
+; AVX512:       # %bb.0: # %bb1
+; AVX512-NEXT:    vpmaddwd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0 # [4096,1,4096,1,4096,1,4096,1,4096,1,4096,1,4096,1,4096,1]
+; AVX512-NEXT:    retq
 bb1:
   %0 = sext <16 x i16> %v to <16 x i32>
   %1 = shl nsw <16 x i32> %0, <i32 12, i32 0, i32 12, i32 0, i32 12, i32 0, i32 12, i32 0, i32 12, i32 0, i32 12, i32 0, i32 12, i32 0, i32 12, i32 0>
@@ -415,21 +468,21 @@ define <8 x i32> @combine_with_shl_overflow(<16 x i16> %v) {
 ; SSE2-NEXT:    paddd %xmm4, %xmm0
 ; SSE2-NEXT:    retq
 ;
-; SSE41-LABEL: combine_with_shl_overflow:
-; SSE41:       # %bb.0: # %bb1
-; SSE41-NEXT:    pxor %xmm2, %xmm2
-; SSE41-NEXT:    pxor %xmm4, %xmm4
-; SSE41-NEXT:    punpckhwd {{.*#+}} xmm4 = xmm4[4],xmm1[4],xmm4[5],xmm1[5],xmm4[6],xmm1[6],xmm4[7],xmm1[7]
-; SSE41-NEXT:    pxor %xmm3, %xmm3
-; SSE41-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1],xmm3[2],xmm1[2],xmm3[3],xmm1[3]
-; SSE41-NEXT:    phaddd %xmm4, %xmm3
-; SSE41-NEXT:    pxor %xmm1, %xmm1
-; SSE41-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
-; SSE41-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; SSE41-NEXT:    phaddd %xmm1, %xmm2
-; SSE41-NEXT:    movdqa %xmm2, %xmm0
-; SSE41-NEXT:    movdqa %xmm3, %xmm1
-; SSE41-NEXT:    retq
+; SSE42-LABEL: combine_with_shl_overflow:
+; SSE42:       # %bb.0: # %bb1
+; SSE42-NEXT:    pxor %xmm3, %xmm3
+; SSE42-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm1[4],xmm3[5],xmm1[5],xmm3[6],xmm1[6],xmm3[7],xmm1[7]
+; SSE42-NEXT:    pxor %xmm2, %xmm2
+; SSE42-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
+; SSE42-NEXT:    phaddd %xmm3, %xmm2
+; SSE42-NEXT:    pxor %xmm1, %xmm1
+; SSE42-NEXT:    pxor %xmm3, %xmm3
+; SSE42-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
+; SSE42-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; SSE42-NEXT:    phaddd %xmm3, %xmm1
+; SSE42-NEXT:    movdqa %xmm1, %xmm0
+; SSE42-NEXT:    movdqa %xmm2, %xmm1
+; SSE42-NEXT:    retq
 ;
 ; AVX1-LABEL: combine_with_shl_overflow:
 ; AVX1:       # %bb.0: # %bb1
@@ -456,6 +509,15 @@ define <8 x i32> @combine_with_shl_overflow(<16 x i16> %v) {
 ; AVX2-NEXT:    vphaddd %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
 ; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: combine_with_shl_overflow:
+; AVX512:       # %bb.0: # %bb1
+; AVX512-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
+; AVX512-NEXT:    vpslld $16, %zmm0, %zmm0
+; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
+; AVX512-NEXT:    vphaddd %ymm1, %ymm0, %ymm0
+; AVX512-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
+; AVX512-NEXT:    retq
 bb1:
   %0 = sext <16 x i16> %v to <16 x i32>
   %1 = shl nsw <16 x i32> %0, splat (i32 16)
@@ -464,3 +526,5 @@ bb1:
   %4 = add <8 x i32> %2, %3
   ret <8 x i32> %4
 }
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; CHECK: {{.*}}



More information about the llvm-commits mailing list