[DAG] Add ISD::FP_TO_SINT_SAT/FP_TO_UINT_SAT handling to SelectionDAG::canCreateUndefOrPoison (#154244)

Related to https://github.com/llvm/llvm-project/issues/153366
This commit is contained in:
Ye Tian 2025-08-19 09:45:11 +08:00 committed by GitHub
parent 1eb5b18a04
commit db843e5d09
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 85 additions and 0 deletions

View File

@ -5748,6 +5748,8 @@ bool SelectionDAG::canCreateUndefOrPoison(SDValue Op, const APInt &DemandedElts,
case ISD::FMA:
case ISD::FMAD:
case ISD::FP_EXTEND:
case ISD::FP_TO_SINT_SAT:
case ISD::FP_TO_UINT_SAT:
// No poison except from flags (which is handled above)
return false;

View File

@ -57,6 +57,52 @@ define <4 x i1> @test_signed_v4i1_v4f32(<4 x float> %f) nounwind {
ret <4 x i1> %x
}
define <4 x i1> @test_freeze_signed_v4i1_v4f32(<4 x float> %f) nounwind {
; CHECK-LABEL: test_freeze_signed_v4i1_v4f32:
; CHECK: # %bb.0:
; CHECK-NEXT: movaps %xmm0, %xmm1
; CHECK-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,3],xmm0[3,3]
; CHECK-NEXT: movss {{.*#+}} xmm2 = [-1.0E+0,0.0E+0,0.0E+0,0.0E+0]
; CHECK-NEXT: xorl %eax, %eax
; CHECK-NEXT: ucomiss %xmm1, %xmm1
; CHECK-NEXT: maxss %xmm2, %xmm1
; CHECK-NEXT: xorps %xmm3, %xmm3
; CHECK-NEXT: minss %xmm3, %xmm1
; CHECK-NEXT: cvttss2si %xmm1, %ecx
; CHECK-NEXT: cmovpl %eax, %ecx
; CHECK-NEXT: movd %ecx, %xmm1
; CHECK-NEXT: movaps %xmm0, %xmm4
; CHECK-NEXT: unpckhpd {{.*#+}} xmm4 = xmm4[1],xmm0[1]
; CHECK-NEXT: ucomiss %xmm4, %xmm4
; CHECK-NEXT: maxss %xmm2, %xmm4
; CHECK-NEXT: minss %xmm3, %xmm4
; CHECK-NEXT: cvttss2si %xmm4, %ecx
; CHECK-NEXT: cmovpl %eax, %ecx
; CHECK-NEXT: movd %ecx, %xmm4
; CHECK-NEXT: punpckldq {{.*#+}} xmm4 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
; CHECK-NEXT: movaps %xmm0, %xmm1
; CHECK-NEXT: maxss %xmm2, %xmm1
; CHECK-NEXT: minss %xmm3, %xmm1
; CHECK-NEXT: cvttss2si %xmm1, %ecx
; CHECK-NEXT: ucomiss %xmm0, %xmm0
; CHECK-NEXT: cmovpl %eax, %ecx
; CHECK-NEXT: movd %ecx, %xmm1
; CHECK-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
; CHECK-NEXT: ucomiss %xmm0, %xmm0
; CHECK-NEXT: maxss %xmm2, %xmm0
; CHECK-NEXT: minss %xmm3, %xmm0
; CHECK-NEXT: cvttss2si %xmm0, %ecx
; CHECK-NEXT: cmovpl %eax, %ecx
; CHECK-NEXT: movd %ecx, %xmm0
; CHECK-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
; CHECK-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm4[0]
; CHECK-NEXT: movdqa %xmm1, %xmm0
; CHECK-NEXT: retq
%x = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> %f)
%y = freeze <4 x i1> %x
ret <4 x i1> %y
}
define <4 x i8> @test_signed_v4i8_v4f32(<4 x float> %f) nounwind {
; CHECK-LABEL: test_signed_v4i8_v4f32:
; CHECK: # %bb.0:

View File

@ -48,6 +48,43 @@ define <4 x i1> @test_unsigned_v4i1_v4f32(<4 x float> %f) nounwind {
ret <4 x i1> %x
}
define <4 x i1> @test_freeze_unsigned_v4i1_v4f32(<4 x float> %f) nounwind {
; CHECK-LABEL: test_freeze_unsigned_v4i1_v4f32:
; CHECK: # %bb.0:
; CHECK-NEXT: movaps %xmm0, %xmm1
; CHECK-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,3],xmm0[3,3]
; CHECK-NEXT: xorps %xmm2, %xmm2
; CHECK-NEXT: maxss %xmm2, %xmm1
; CHECK-NEXT: movss {{.*#+}} xmm3 = [1.0E+0,0.0E+0,0.0E+0,0.0E+0]
; CHECK-NEXT: minss %xmm3, %xmm1
; CHECK-NEXT: cvttss2si %xmm1, %eax
; CHECK-NEXT: movd %eax, %xmm1
; CHECK-NEXT: movaps %xmm0, %xmm4
; CHECK-NEXT: unpckhpd {{.*#+}} xmm4 = xmm4[1],xmm0[1]
; CHECK-NEXT: maxss %xmm2, %xmm4
; CHECK-NEXT: minss %xmm3, %xmm4
; CHECK-NEXT: cvttss2si %xmm4, %eax
; CHECK-NEXT: movd %eax, %xmm4
; CHECK-NEXT: punpckldq {{.*#+}} xmm4 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
; CHECK-NEXT: movaps %xmm0, %xmm1
; CHECK-NEXT: maxss %xmm2, %xmm1
; CHECK-NEXT: minss %xmm3, %xmm1
; CHECK-NEXT: cvttss2si %xmm1, %eax
; CHECK-NEXT: movd %eax, %xmm1
; CHECK-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
; CHECK-NEXT: maxss %xmm2, %xmm0
; CHECK-NEXT: minss %xmm3, %xmm0
; CHECK-NEXT: cvttss2si %xmm0, %eax
; CHECK-NEXT: movd %eax, %xmm0
; CHECK-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
; CHECK-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm4[0]
; CHECK-NEXT: movdqa %xmm1, %xmm0
; CHECK-NEXT: retq
%x = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> %f)
%y = freeze <4 x i1> %x
ret <4 x i1> %y
}
define <4 x i8> @test_unsigned_v4i8_v4f32(<4 x float> %f) nounwind {
; CHECK-LABEL: test_unsigned_v4i8_v4f32:
; CHECK: # %bb.0: