release/20.x: [X86] Ignore NSW when DstSVT is i32 (#131755)

llvm/lib/Target/X86/X86ISelLowering.cpp

@@ -20889,7 +20889,8 @@ static SDValue matchTruncateWithPACK(unsigned &PackOpcode, EVT DstVT,
     return SDValue();
 
   unsigned MinSignBits = NumSrcEltBits - NumPackedSignBits;
-  if (Flags.hasNoSignedWrap() || MinSignBits < NumSignBits) {
+  if ((Flags.hasNoSignedWrap() && DstSVT != MVT::i32) ||
+      MinSignBits < NumSignBits) {
     PackOpcode = X86ISD::PACKSS;
     return In;
   }

llvm/test/CodeGen/X86/vector-trunc-nowrap.ll

@@ -1592,3 +1592,91 @@ entry:
   %1 = bitcast <8 x i8> %0 to i64
   ret i64 %1
 }
+
+define void @foo(<4 x i64> %a, <4 x i64> %b, ptr %p) "min-legal-vector-width"="256" "prefer-vector-width"="256" {
+; SSE-LABEL: foo:
+; SSE:       # %bb.0: # %entry
+; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,2],xmm3[0,2]
+; SSE-NEXT:    movaps %xmm2, 16(%rdi)
+; SSE-NEXT:    movaps %xmm0, (%rdi)
+; SSE-NEXT:    retq
+;
+; AVX1-LABEL: foo:
+; AVX1:       # %bb.0: # %entry
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
+; AVX1-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
+; AVX1-NEXT:    vmovaps %xmm1, 16(%rdi)
+; AVX1-NEXT:    vmovaps %xmm0, (%rdi)
+; AVX1-NEXT:    vzeroupper
+; AVX1-NEXT:    retq
+;
+; AVX2-SLOW-LABEL: foo:
+; AVX2-SLOW:       # %bb.0: # %entry
+; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm2
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
+; AVX2-SLOW-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX2-SLOW-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
+; AVX2-SLOW-NEXT:    vmovaps %xmm1, 16(%rdi)
+; AVX2-SLOW-NEXT:    vmovaps %xmm0, (%rdi)
+; AVX2-SLOW-NEXT:    vzeroupper
+; AVX2-SLOW-NEXT:    retq
+;
+; AVX2-FAST-ALL-LABEL: foo:
+; AVX2-FAST-ALL:       # %bb.0: # %entry
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm2 = [0,2,4,6,4,6,6,7]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vmovaps %xmm1, 16(%rdi)
+; AVX2-FAST-ALL-NEXT:    vmovaps %xmm0, (%rdi)
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: foo:
+; AVX2-FAST-PERLANE:       # %bb.0: # %entry
+; AVX2-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm0, %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vmovaps %xmm1, 16(%rdi)
+; AVX2-FAST-PERLANE-NEXT:    vmovaps %xmm0, (%rdi)
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
+;
+; AVX512F-LABEL: foo:
+; AVX512F:       # %bb.0: # %entry
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
+; AVX512F-NEXT:    vpmovqd %zmm0, (%rdi)
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: foo:
+; AVX512VL:       # %bb.0: # %entry
+; AVX512VL-NEXT:    vpmovqd %ymm1, 16(%rdi)
+; AVX512VL-NEXT:    vpmovqd %ymm0, (%rdi)
+; AVX512VL-NEXT:    vzeroupper
+; AVX512VL-NEXT:    retq
+;
+; AVX512BW-LABEL: foo:
+; AVX512BW:       # %bb.0: # %entry
+; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512BW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpmovqd %zmm0, (%rdi)
+; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    retq
+;
+; AVX512BWVL-LABEL: foo:
+; AVX512BWVL:       # %bb.0: # %entry
+; AVX512BWVL-NEXT:    vpmovqd %ymm1, 16(%rdi)
+; AVX512BWVL-NEXT:    vpmovqd %ymm0, (%rdi)
+; AVX512BWVL-NEXT:    vzeroupper
+; AVX512BWVL-NEXT:    retq
+entry:
+  %0 = shufflevector <4 x i64> %a, <4 x i64> %b, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+  %1 = trunc nsw <8 x i64> %0 to <8 x i32>
+  store <8 x i32> %1, ptr %p, align 16
+  ret void
+}