[x86] allow narrowing of vector UINT_TO_FP

As discussed in:
D56864
D58197

Always use the narrow (128-bit) instruction when possible.
We already had the signed int version of this transform.

llvm-svn: 354675

Author: rotateright

llvm/lib/Target/X86/X86ISelLowering.cpp

@@ -42016,6 +42016,11 @@ static SDValue combineExtractSubvector(SDNode *N, SelectionDAG &DAG,
           InVec.getOperand(0).getValueType() == MVT::v4i32) {
         return DAG.getNode(X86ISD::CVTSI2P, SDLoc(N), VT, InVec.getOperand(0));
       }
+      // v2f64 CVTUDQ2PD(v4i32).
+      if (InOpcode == ISD::UINT_TO_FP &&
+          InVec.getOperand(0).getValueType() == MVT::v4i32) {
+        return DAG.getNode(X86ISD::CVTUI2P, SDLoc(N), VT, InVec.getOperand(0));
+      }
       // v2f64 CVTPS2PD(v4f32).
       if (InOpcode == ISD::FP_EXTEND &&
           InVec.getOperand(0).getValueType() == MVT::v4f32) {

llvm/test/CodeGen/X86/vec_int_to_fp-widen.ll

@@ -684,9 +684,7 @@ define <2 x double> @uitofp_4i32_to_2f64(<4 x i32> %a) {
 ;
 ; AVX512VL-LABEL: uitofp_4i32_to_2f64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vcvtudq2pd %xmm0, %ymm0
-; AVX512VL-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
-; AVX512VL-NEXT:    vzeroupper
+; AVX512VL-NEXT:    vcvtudq2pd %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: uitofp_4i32_to_2f64:
@@ -699,9 +697,7 @@ define <2 x double> @uitofp_4i32_to_2f64(<4 x i32> %a) {
 ;
 ; AVX512VLDQ-LABEL: uitofp_4i32_to_2f64:
 ; AVX512VLDQ:       # %bb.0:
-; AVX512VLDQ-NEXT:    vcvtudq2pd %xmm0, %ymm0
-; AVX512VLDQ-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
-; AVX512VLDQ-NEXT:    vzeroupper
+; AVX512VLDQ-NEXT:    vcvtudq2pd %xmm0, %xmm0
 ; AVX512VLDQ-NEXT:    retq
   %cvt = uitofp <4 x i32> %a to <4 x double>
   %shuf = shufflevector <4 x double> %cvt, <4 x double> undef, <2 x i32> <i32 0, i32 1>

llvm/test/CodeGen/X86/vec_int_to_fp.ll

@@ -684,9 +684,7 @@ define <2 x double> @uitofp_4i32_to_2f64(<4 x i32> %a) {
 ;
 ; AVX512VL-LABEL: uitofp_4i32_to_2f64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vcvtudq2pd %xmm0, %ymm0
-; AVX512VL-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
-; AVX512VL-NEXT:    vzeroupper
+; AVX512VL-NEXT:    vcvtudq2pd %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: uitofp_4i32_to_2f64:
@@ -699,9 +697,7 @@ define <2 x double> @uitofp_4i32_to_2f64(<4 x i32> %a) {
 ;
 ; AVX512VLDQ-LABEL: uitofp_4i32_to_2f64:
 ; AVX512VLDQ:       # %bb.0:
-; AVX512VLDQ-NEXT:    vcvtudq2pd %xmm0, %ymm0
-; AVX512VLDQ-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
-; AVX512VLDQ-NEXT:    vzeroupper
+; AVX512VLDQ-NEXT:    vcvtudq2pd %xmm0, %xmm0
 ; AVX512VLDQ-NEXT:    retq
   %cvt = uitofp <4 x i32> %a to <4 x double>
   %shuf = shufflevector <4 x double> %cvt, <4 x double> undef, <2 x i32> <i32 0, i32 1>
@@ -5699,9 +5695,7 @@ define double @extract0_uitofp_v4i32_f64(<4 x i32> %x) nounwind {
 ;
 ; AVX512VL-LABEL: extract0_uitofp_v4i32_f64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vcvtudq2pd %xmm0, %ymm0
-; AVX512VL-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
-; AVX512VL-NEXT:    vzeroupper
+; AVX512VL-NEXT:    vcvtudq2pd %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: extract0_uitofp_v4i32_f64:
@@ -5714,9 +5708,7 @@ define double @extract0_uitofp_v4i32_f64(<4 x i32> %x) nounwind {
 ;
 ; AVX512VLDQ-LABEL: extract0_uitofp_v4i32_f64:
 ; AVX512VLDQ:       # %bb.0:
-; AVX512VLDQ-NEXT:    vcvtudq2pd %xmm0, %ymm0
-; AVX512VLDQ-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
-; AVX512VLDQ-NEXT:    vzeroupper
+; AVX512VLDQ-NEXT:    vcvtudq2pd %xmm0, %xmm0
 ; AVX512VLDQ-NEXT:    retq
   %e = extractelement <4 x i32> %x, i32 0
   %r = uitofp i32 %e to double
@@ -5855,9 +5847,7 @@ define double @extract3_uitofp_v4i32_f64(<4 x i32> %x) nounwind {
 ; AVX512VL-LABEL: extract3_uitofp_v4i32_f64:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512VL-NEXT:    vcvtudq2pd %xmm0, %ymm0
-; AVX512VL-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
-; AVX512VL-NEXT:    vzeroupper
+; AVX512VL-NEXT:    vcvtudq2pd %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: extract3_uitofp_v4i32_f64:
@@ -5871,9 +5861,7 @@ define double @extract3_uitofp_v4i32_f64(<4 x i32> %x) nounwind {
 ; AVX512VLDQ-LABEL: extract3_uitofp_v4i32_f64:
 ; AVX512VLDQ:       # %bb.0:
 ; AVX512VLDQ-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512VLDQ-NEXT:    vcvtudq2pd %xmm0, %ymm0
-; AVX512VLDQ-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
-; AVX512VLDQ-NEXT:    vzeroupper
+; AVX512VLDQ-NEXT:    vcvtudq2pd %xmm0, %xmm0
 ; AVX512VLDQ-NEXT:    retq
   %e = extractelement <4 x i32> %x, i32 3
   %r = uitofp i32 %e to double