bool FloatDomain = MaskVT.isFloatingPoint() ||
(!Subtarget.hasAVX2() && MaskVT.is256BitVector());
- // Match a 128-bit integer vector against a VZEXT_MOVL (MOVQ) instruction.
- if (!FloatDomain && MaskVT.is128BitVector() &&
+ // Match a 128-bit vector against a VZEXT_MOVL instruction.
+ if (MaskVT.is128BitVector() && Subtarget.hasSSE2() &&
isTargetShuffleEquivalent(Mask, {0, SM_SentinelZero})) {
Shuffle = X86ISD::VZEXT_MOVL;
- ShuffleVT = MVT::v2i64;
+ ShuffleVT = MaskVT;
return true;
}
; SSE41-LABEL: combine_scalar_load_with_blend_with_zero:
; SSE41: # BB#0:
; SSE41-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
-; SSE41-NEXT: xorpd %xmm1, %xmm1
-; SSE41-NEXT: blendpd {{.*#+}} xmm1 = xmm0[0],xmm1[1]
-; SSE41-NEXT: movapd %xmm1, (%rsi)
+; SSE41-NEXT: movapd %xmm0, (%rsi)
; SSE41-NEXT: retq
;
; AVX-LABEL: combine_scalar_load_with_blend_with_zero:
; AVX: # BB#0:
; AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
-; AVX-NEXT: vxorpd %xmm1, %xmm1, %xmm1
-; AVX-NEXT: vblendpd {{.*#+}} xmm0 = xmm0[0],xmm1[1]
; AVX-NEXT: vmovapd %xmm0, (%rsi)
; AVX-NEXT: retq
%1 = load double, double* %a0, align 8