[llvm] 2137859 - [X86] Add PR34666 redundant broadcast test case

[Simon Pilgrim via llvm-commits]

Author: Simon Pilgrim
Date: 2022-03-21T16:10:06Z
New Revision: 21378593fbbac3a184fd3a1ab6d059cc94490f41

URL: https://github.com/llvm/llvm-project/commit/21378593fbbac3a184fd3a1ab6d059cc94490f41
DIFF: https://github.com/llvm/llvm-project/commit/21378593fbbac3a184fd3a1ab6d059cc94490f41.diff

LOG: [X86] Add PR34666 redundant broadcast test case

Added: 
    llvm/test/CodeGen/X86/avx512-broadcast-arith.ll

Modified: 
    

Removed: 
    


################################################################################
diff  --git a/llvm/test/CodeGen/X86/avx512-broadcast-arith.ll b/llvm/test/CodeGen/X86/avx512-broadcast-arith.ll
new file mode 100644
index 0000000000000..22fff49103c22
--- /dev/null
+++ b/llvm/test/CodeGen/X86/avx512-broadcast-arith.ll
@@ -0,0 +1,60 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx512f           | FileCheck %s --check-prefixes=AVX512F
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=AVX512BW
+
+; PR34666
+define <64 x i8> @add_v64i8_broadcasts(<64 x i8> %a0, i64 %a1, i8 %a2) {
+; AVX512F-LABEL: add_v64i8_broadcasts:
+; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    movq %rdi, %rax
+; AVX512F-NEXT:    movl %edi, %ecx
+; AVX512F-NEXT:    kmovw %edi, %k1
+; AVX512F-NEXT:    shrq $32, %rdi
+; AVX512F-NEXT:    shrq $48, %rax
+; AVX512F-NEXT:    shrl $16, %ecx
+; AVX512F-NEXT:    vmovd %esi, %xmm1
+; AVX512F-NEXT:    vpbroadcastb %xmm1, %ymm1
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm1, %zmm1, %zmm1
+; AVX512F-NEXT:    kmovw %ecx, %k2
+; AVX512F-NEXT:    kmovw %eax, %k3
+; AVX512F-NEXT:    kmovw %edi, %k4
+; AVX512F-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k4} {z}
+; AVX512F-NEXT:    vpmovdb %zmm2, %xmm2
+; AVX512F-NEXT:    vpternlogd $255, %zmm3, %zmm3, %zmm3 {%k3} {z}
+; AVX512F-NEXT:    vpmovdb %zmm3, %xmm3
+; AVX512F-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpternlogd $255, %zmm3, %zmm3, %zmm3 {%k1} {z}
+; AVX512F-NEXT:    vpmovdb %zmm3, %xmm3
+; AVX512F-NEXT:    vpternlogd $255, %zmm4, %zmm4, %zmm4 {%k2} {z}
+; AVX512F-NEXT:    vpmovdb %zmm4, %xmm4
+; AVX512F-NEXT:    vinserti128 $1, %xmm4, %ymm3, %ymm3
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm3, %zmm2
+; AVX512F-NEXT:    vpternlogq $216, %zmm2, %zmm1, %zmm0
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm3
+; AVX512F-NEXT:    vpaddb %ymm1, %ymm3, %ymm3
+; AVX512F-NEXT:    vpaddb %ymm1, %ymm0, %ymm0
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm0, %zmm3
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm3, %ymm0
+; AVX512F-NEXT:    vpaddb %ymm1, %ymm0, %ymm0
+; AVX512F-NEXT:    vpaddb %ymm1, %ymm3, %ymm1
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512F-NEXT:    vpternlogq $226, %zmm3, %zmm2, %zmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512BW-LABEL: add_v64i8_broadcasts:
+; AVX512BW:       # %bb.0:
+; AVX512BW-NEXT:    vpbroadcastb %esi, %zmm1
+; AVX512BW-NEXT:    kmovq %rdi, %k1
+; AVX512BW-NEXT:    vpbroadcastb %esi, %zmm0 {%k1}
+; AVX512BW-NEXT:    vpaddb %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpaddb %zmm1, %zmm0, %zmm0 {%k1}
+; AVX512BW-NEXT:    retq
+  %v = insertelement <64 x i8> undef, i8 %a2, i64 0
+  %broadcast = shufflevector <64 x i8> %v, <64 x i8> poison, <64 x i32> zeroinitializer
+  %mask = bitcast i64 %a1 to <64 x i1>
+  %broadcast_mask = select <64 x i1> %mask, <64 x i8> %broadcast, <64 x i8> %a0
+  %broadcast_maskz = select <64 x i1> %mask, <64 x i8> %broadcast, <64 x i8> zeroinitializer
+  %add0 = add <64 x i8> %broadcast_mask, %broadcast
+  %add1 = add <64 x i8> %add0, %broadcast_maskz
+  ret <64 x i8> %add1
+}