[llvm] 05671ce - [X86] Add tests showing failure to push freeze through SSE PMADD nodes

[Simon Pilgrim via llvm-commits]

Author: Simon Pilgrim
Date: 2024-07-22T14:31:11+01:00
New Revision: 05671cec6f736c65b7df29234db828116e6d95dd

URL: https://github.com/llvm/llvm-project/commit/05671cec6f736c65b7df29234db828116e6d95dd
DIFF: https://github.com/llvm/llvm-project/commit/05671cec6f736c65b7df29234db828116e6d95dd.diff

LOG: [X86] Add tests showing failure to push freeze through SSE PMADD nodes

PMADD guarantee inbounds/saturated ext-multiply-add results

Test to help with regression identified on #84924

Added: 
    

Modified: 
    llvm/test/CodeGen/X86/combine-pmadd.ll

Removed: 
    


################################################################################
diff  --git a/llvm/test/CodeGen/X86/combine-pmadd.ll b/llvm/test/CodeGen/X86/combine-pmadd.ll
index 565d9ef6eb3e6..8ba5a25678122 100644
--- a/llvm/test/CodeGen/X86/combine-pmadd.ll
+++ b/llvm/test/CodeGen/X86/combine-pmadd.ll
@@ -63,6 +63,37 @@ define <8 x i32> @combine_pmaddwd_concat(<8 x i16> %a0, <8 x i16> %a1, <8 x i16>
   ret <8 x i32> %3
 }
 
+define <8 x i32> @combine_pmaddwd_concat_freeze(<8 x i16> %a0, <8 x i16> %a1) {
+; SSE-LABEL: combine_pmaddwd_concat_freeze:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pmovsxbw {{.*#+}} 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:
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm2 = [1,1,1,1,1,1,1,1]
+; AVX1-NEXT:    vpmaddwd %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpmaddwd %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: combine_pmaddwd_concat_freeze:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpbroadcastw {{.*#+}} xmm2 = [1,1,1,1,1,1,1,1]
+; AVX2-NEXT:    vpmaddwd %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpmaddwd %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-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
+  %fhi = freeze <4 x i32> %hi
+  %res = shufflevector <4 x i32> %flo, <4 x i32> %fhi, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+  ret <8 x i32> %res
+}
+
 define <4 x i32> @combine_pmaddwd_demandedelts(<8 x i16> %a0, <8 x i16> %a1) {
 ; SSE-LABEL: combine_pmaddwd_demandedelts:
 ; SSE:       # %bb.0:
@@ -178,6 +209,37 @@ define <16 x i16> @combine_pmaddubsw_concat(<16 x i8> %a0, <16 x i8> %a1, <16 x
   ret <16 x i16> %3
 }
 
+define <16 x i16> @combine_pmaddubsw_concat_freeze(<16 x i8> %a0, <16 x i8> %a1) {
+; SSE-LABEL: combine_pmaddubsw_concat_freeze:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
+; SSE-NEXT:    pmaddubsw %xmm2, %xmm0
+; SSE-NEXT:    pmaddubsw %xmm2, %xmm1
+; SSE-NEXT:    retq
+;
+; AVX1-LABEL: combine_pmaddubsw_concat_freeze:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm2 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
+; AVX1-NEXT:    vpmaddubsw %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpmaddubsw %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: combine_pmaddubsw_concat_freeze:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpbroadcastb {{.*#+}} xmm2 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
+; AVX2-NEXT:    vpmaddubsw %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpmaddubsw %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-NEXT:    retq
+  %lo = call <8 x i16> @llvm.x86.ssse3.pmadd.ub.sw.128(<16 x i8> %a0, <16 x i8> <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>)
+  %hi = call <8 x i16> @llvm.x86.ssse3.pmadd.ub.sw.128(<16 x i8> %a1, <16 x i8> <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>)
+  %flo = freeze <8 x i16> %lo
+  %fhi = freeze <8 x i16> %hi
+  %res = shufflevector <8 x i16> %flo, <8 x i16> %fhi, <16 x i32> <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>
+  ret <16 x i16> %res
+}
+
 define <8 x i16> @combine_pmaddubsw_demandedelts(<16 x i8> %a0, <16 x i8> %a1) {
 ; SSE-LABEL: combine_pmaddubsw_demandedelts:
 ; SSE:       # %bb.0: