[llvm-branch-commits] [llvm] 496b174 - [X86] Add tests for incorrectly optimizing out shuffle used in `movmsk`; PR67287

[Tobias Hieta via llvm-branch-commits]

Author: Noah Goldstein
Date: 2023-10-16T10:15:06+02:00
New Revision: 496b174053bd77d2a47de16dbf86146109979590

URL: https://github.com/llvm/llvm-project/commit/496b174053bd77d2a47de16dbf86146109979590
DIFF: https://github.com/llvm/llvm-project/commit/496b174053bd77d2a47de16dbf86146109979590.diff

LOG: [X86] Add tests for incorrectly optimizing out shuffle used in `movmsk`; PR67287

(cherry picked from commit 65a576e27be814bd23f39b31a8074e9850b0fe26)

Added: 
    

Modified: 
    llvm/test/CodeGen/X86/movmsk-cmp.ll

Removed: 
    


################################################################################
diff  --git a/llvm/test/CodeGen/X86/movmsk-cmp.ll b/llvm/test/CodeGen/X86/movmsk-cmp.ll
index 6ed44771f703e56..f7ba49ce0e12732 100644
--- a/llvm/test/CodeGen/X86/movmsk-cmp.ll
+++ b/llvm/test/CodeGen/X86/movmsk-cmp.ll
@@ -4454,3 +4454,140 @@ define i32 @PR39665_c_ray_opt(<2 x double> %x, <2 x double> %y) {
   %r = select i1 %u, i32 42, i32 99
   ret i32 %r
 }
+
+define i32 @pr67287(<2 x i64> %broadcast.splatinsert25) {
+; SSE2-LABEL: pr67287:
+; SSE2:       # %bb.0: # %entry
+; SSE2-NEXT:    movl $3, %eax
+; SSE2-NEXT:    testl %eax, %eax
+; SSE2-NEXT:    jne .LBB97_2
+; SSE2-NEXT:  # %bb.1: # %entry
+; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE2-NEXT:    pxor %xmm1, %xmm1
+; SSE2-NEXT:    pcmpeqd %xmm0, %xmm1
+; SSE2-NEXT:    movd %xmm1, %eax
+; SSE2-NEXT:    testb $1, %al
+; SSE2-NEXT:    jne .LBB97_2
+; SSE2-NEXT:  # %bb.3: # %middle.block
+; SSE2-NEXT:    xorl %eax, %eax
+; SSE2-NEXT:    retq
+; SSE2-NEXT:  .LBB97_2:
+; SSE2-NEXT:    movw $0, 0
+; SSE2-NEXT:    xorl %eax, %eax
+; SSE2-NEXT:    retq
+;
+; SSE41-LABEL: pr67287:
+; SSE41:       # %bb.0: # %entry
+; SSE41-NEXT:    pxor %xmm1, %xmm1
+; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
+; SSE41-NEXT:    pcmpeqq %xmm1, %xmm0
+; SSE41-NEXT:    movmskpd %xmm0, %eax
+; SSE41-NEXT:    testl %eax, %eax
+; SSE41-NEXT:    jne .LBB97_2
+; SSE41-NEXT:  # %bb.1: # %entry
+; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    testb $1, %al
+; SSE41-NEXT:    jne .LBB97_2
+; SSE41-NEXT:  # %bb.3: # %middle.block
+; SSE41-NEXT:    xorl %eax, %eax
+; SSE41-NEXT:    retq
+; SSE41-NEXT:  .LBB97_2:
+; SSE41-NEXT:    movw $0, 0
+; SSE41-NEXT:    xorl %eax, %eax
+; SSE41-NEXT:    retq
+;
+; AVX1-LABEL: pr67287:
+; AVX1:       # %bb.0: # %entry
+; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
+; AVX1-NEXT:    vpcmpeqq %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vtestpd %xmm0, %xmm0
+; AVX1-NEXT:    jne .LBB97_2
+; AVX1-NEXT:  # %bb.1: # %entry
+; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    testb $1, %al
+; AVX1-NEXT:    jne .LBB97_2
+; AVX1-NEXT:  # %bb.3: # %middle.block
+; AVX1-NEXT:    xorl %eax, %eax
+; AVX1-NEXT:    retq
+; AVX1-NEXT:  .LBB97_2:
+; AVX1-NEXT:    movw $0, 0
+; AVX1-NEXT:    xorl %eax, %eax
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: pr67287:
+; AVX2:       # %bb.0: # %entry
+; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
+; AVX2-NEXT:    vpcmpeqq %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vtestpd %xmm0, %xmm0
+; AVX2-NEXT:    jne .LBB97_2
+; AVX2-NEXT:  # %bb.1: # %entry
+; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    testb $1, %al
+; AVX2-NEXT:    jne .LBB97_2
+; AVX2-NEXT:  # %bb.3: # %middle.block
+; AVX2-NEXT:    xorl %eax, %eax
+; AVX2-NEXT:    retq
+; AVX2-NEXT:  .LBB97_2:
+; AVX2-NEXT:    movw $0, 0
+; AVX2-NEXT:    xorl %eax, %eax
+; AVX2-NEXT:    retq
+;
+; KNL-LABEL: pr67287:
+; KNL:       # %bb.0: # %entry
+; KNL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; KNL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
+; KNL-NEXT:    vptestnmq %zmm0, %zmm0, %k0
+; KNL-NEXT:    kmovw %k0, %eax
+; KNL-NEXT:    testb $3, %al
+; KNL-NEXT:    jne .LBB97_2
+; KNL-NEXT:  # %bb.1: # %entry
+; KNL-NEXT:    kmovw %k0, %eax
+; KNL-NEXT:    testb $1, %al
+; KNL-NEXT:    jne .LBB97_2
+; KNL-NEXT:  # %bb.3: # %middle.block
+; KNL-NEXT:    xorl %eax, %eax
+; KNL-NEXT:    vzeroupper
+; KNL-NEXT:    retq
+; KNL-NEXT:  .LBB97_2:
+; KNL-NEXT:    movw $0, 0
+; KNL-NEXT:    xorl %eax, %eax
+; KNL-NEXT:    vzeroupper
+; KNL-NEXT:    retq
+;
+; SKX-LABEL: pr67287:
+; SKX:       # %bb.0: # %entry
+; SKX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; SKX-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
+; SKX-NEXT:    vptestnmq %xmm0, %xmm0, %k0
+; SKX-NEXT:    kortestb %k0, %k0
+; SKX-NEXT:    jne .LBB97_2
+; SKX-NEXT:  # %bb.1: # %entry
+; SKX-NEXT:    kmovd %k0, %eax
+; SKX-NEXT:    testb $1, %al
+; SKX-NEXT:    jne .LBB97_2
+; SKX-NEXT:  # %bb.3: # %middle.block
+; SKX-NEXT:    xorl %eax, %eax
+; SKX-NEXT:    retq
+; SKX-NEXT:  .LBB97_2:
+; SKX-NEXT:    movw $0, 0
+; SKX-NEXT:    xorl %eax, %eax
+; SKX-NEXT:    retq
+entry:
+  %0 = and <2 x i64> %broadcast.splatinsert25, <i64 4294967295, i64 4294967295>
+  %1 = icmp eq <2 x i64> %0, zeroinitializer
+  %shift = shufflevector <2 x i1> %1, <2 x i1> zeroinitializer, <2 x i32> <i32 1, i32 poison>
+  %2 = or <2 x i1> %1, %shift
+  %3 = extractelement <2 x i1> %2, i64 0
+  %4 = extractelement <2 x i1> %1, i64 0
+  %5 = or i1 %3, %4
+  br i1 %5, label %6, label %middle.block
+
+6:                                                ; preds = %entry
+  store i16 0, ptr null, align 2
+  br label %middle.block
+
+middle.block:                                     ; preds = %6, %entry
+  ret i32 0
+}