[llvm-branch-commits] [llvm] [X86] Mark vXi64 int to fp undesirable without AVX512DQ (PR #242914)
Dmitry Sidorov via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Sun Nov 15 16:08:30 PST 2026
https://github.com/MrSidims created https://github.com/llvm/llvm-project/pull/242914
Without avx512dq there is no vector conversion from i64 elements to f32
or f64, so widening the conversion of a source truncated with nsw or nuw
back to i64 makes it slower.
avx512fp16 converts i64 elements to f16 directly, so the widening folds
now pass the node and the check sees the result type.
Assisted-by: Claude Code Opus 5.5
---
<sub>Stack created with <a href="https://github.com/github/gh-stack">GitHub Stacks CLI</a> • <a href="https://gh.io/stacks-feedback">Give Feedback 💬</a></sub>
>From a45fa4206873b8ef7992cff1e3673f2f3be230d6 Mon Sep 17 00:00:00 2001
From: Dmitry Sidorov <Dmitry.Sidorov at amd.com>
Date: Sun, 15 Nov 2026 23:17:42 +0100
Subject: [PATCH] [X86] Mark vXi64 int to fp undesirable without AVX512DQ
Without avx512dq there is no vector conversion from i64 elements to f32
or f64, so widening the conversion of a source truncated with nsw or nuw
back to i64 makes it slower.
avx512fp16 converts i64 elements to f16 directly, so the widening folds
now pass the node and the check sees the result type.
Assisted-by: Claude Code Opus 5.5
---
llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp | 6 +-
llvm/lib/Target/X86/X86ISelLowering.cpp | 10 +
llvm/lib/Target/X86/X86ISelLowering.h | 1 +
.../CodeGen/X86/avx512fp16-int-to-fp-trunc.ll | 86 +-------
llvm/test/CodeGen/X86/combine-sitofp.ll | 162 +++------------
llvm/test/CodeGen/X86/combine-uitofp.ll | 191 ++++++------------
6 files changed, 118 insertions(+), 338 deletions(-)
diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index f10658ad99abb2..9b598e92f9eab4 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -20942,8 +20942,7 @@ SDValue DAGCombiner::visitSINT_TO_FP(SDNode *N) {
// fold (sint_to_fp (trunc nsw x)) -> (sint_to_fp x)
if (N0.getOpcode() == ISD::TRUNCATE && N0->getFlags().hasNoSignedWrap() &&
- TLI.isTypeDesirableForOp(ISD::SINT_TO_FP,
- N0.getOperand(0).getValueType()))
+ TLI.isTypeDesirableForOp(N, N0.getOperand(0).getValueType()))
return DAG.getNode(ISD::SINT_TO_FP, DL, VT, N0.getOperand(0));
return SDValue();
@@ -21012,8 +21011,7 @@ SDValue DAGCombiner::visitUINT_TO_FP(SDNode *N) {
// fold (uint_to_fp (trunc nuw x)) -> (uint_to_fp x)
if (N0.getOpcode() == ISD::TRUNCATE && N0->getFlags().hasNoUnsignedWrap() &&
- TLI.isTypeDesirableForOp(ISD::UINT_TO_FP,
- N0.getOperand(0).getValueType()))
+ TLI.isTypeDesirableForOp(N, N0.getOperand(0).getValueType()))
return DAG.getNode(ISD::UINT_TO_FP, DL, VT, N0.getOperand(0));
return SDValue();
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 7b70106f64049c..9ccb435e2f2fdb 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -64658,6 +64658,16 @@ bool X86TargetLowering::isTypeDesirableForOp(unsigned Opc, EVT VT) const {
return true;
}
+bool X86TargetLowering::isTypeDesirableForOp(SDNode *N, EVT VT) const {
+ unsigned Opc = N->getOpcode();
+ // There are no vXi64 to f32 or f64 conversions without AVX512DQ.
+ if ((Opc == ISD::SINT_TO_FP || Opc == ISD::UINT_TO_FP) &&
+ VT.isVectorOf(MVT::i64) && !Subtarget.hasDQI() &&
+ !(Subtarget.hasFP16() && N->getValueType(0).isVectorOf(MVT::f16)))
+ return false;
+ return isTypeDesirableForOp(Opc, VT);
+}
+
SDValue X86TargetLowering::expandIndirectJTBranch(const SDLoc &dl,
SDValue Value, SDValue Addr,
int JTI,
diff --git a/llvm/lib/Target/X86/X86ISelLowering.h b/llvm/lib/Target/X86/X86ISelLowering.h
index 8565fedbbb3089..8837cb73b27740 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.h
+++ b/llvm/lib/Target/X86/X86ISelLowering.h
@@ -197,6 +197,7 @@ namespace llvm {
/// given node type. e.g. On x86 i16 is legal, but undesirable since i16
/// instruction encodings are longer and some i16 instructions are slow.
bool isTypeDesirableForOp(unsigned Opc, EVT VT) const override;
+ bool isTypeDesirableForOp(SDNode *N, EVT VT) const override;
/// Return true if the target has native support for the
/// specified value type and it is 'desirable' to use the type. e.g. On x86
diff --git a/llvm/test/CodeGen/X86/avx512fp16-int-to-fp-trunc.ll b/llvm/test/CodeGen/X86/avx512fp16-int-to-fp-trunc.ll
index 96a26bf2f25db4..cc918b3e94de4a 100644
--- a/llvm/test/CodeGen/X86/avx512fp16-int-to-fp-trunc.ll
+++ b/llvm/test/CodeGen/X86/avx512fp16-int-to-fp-trunc.ll
@@ -206,62 +206,14 @@ define <8 x half> @uitofp_trunc_v8i64_v8f16(<8 x i64> %a0) {
define <8 x float> @sitofp_trunc_nsw_v8i64_v8f32(<8 x i64> %a0) {
; NOVL-LABEL: sitofp_trunc_nsw_v8i64_v8f32:
; NOVL: # %bb.0:
-; NOVL-NEXT: vextracti32x4 $2, %zmm0, %xmm1
-; NOVL-NEXT: vpextrq $1, %xmm1, %rax
-; NOVL-NEXT: vcvtsi2ss %rax, %xmm15, %xmm2
-; NOVL-NEXT: vmovq %xmm1, %rax
-; NOVL-NEXT: vcvtsi2ss %rax, %xmm15, %xmm1
-; NOVL-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[2,3]
-; NOVL-NEXT: vextracti32x4 $3, %zmm0, %xmm2
-; NOVL-NEXT: vmovq %xmm2, %rax
-; NOVL-NEXT: vcvtsi2ss %rax, %xmm15, %xmm3
-; NOVL-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1],xmm3[0],xmm1[3]
-; NOVL-NEXT: vpextrq $1, %xmm2, %rax
-; NOVL-NEXT: vcvtsi2ss %rax, %xmm15, %xmm2
-; NOVL-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1,2],xmm2[0]
-; NOVL-NEXT: vpextrq $1, %xmm0, %rax
-; NOVL-NEXT: vcvtsi2ss %rax, %xmm15, %xmm2
-; NOVL-NEXT: vmovq %xmm0, %rax
-; NOVL-NEXT: vcvtsi2ss %rax, %xmm15, %xmm3
-; NOVL-NEXT: vinsertps {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[2,3]
-; NOVL-NEXT: vextracti128 $1, %ymm0, %xmm0
-; NOVL-NEXT: vmovq %xmm0, %rax
-; NOVL-NEXT: vcvtsi2ss %rax, %xmm15, %xmm3
-; NOVL-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1],xmm3[0],xmm2[3]
-; NOVL-NEXT: vpextrq $1, %xmm0, %rax
-; NOVL-NEXT: vcvtsi2ss %rax, %xmm15, %xmm0
-; NOVL-NEXT: vinsertps {{.*#+}} xmm0 = xmm2[0,1,2],xmm0[0]
-; NOVL-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; NOVL-NEXT: vpmovqd %zmm0, %ymm0
+; NOVL-NEXT: vcvtdq2ps %ymm0, %ymm0
; NOVL-NEXT: retq
;
; VLNODQ-LABEL: sitofp_trunc_nsw_v8i64_v8f32:
; VLNODQ: # %bb.0:
-; VLNODQ-NEXT: vextracti32x4 $2, %zmm0, %xmm1
-; VLNODQ-NEXT: vpextrq $1, %xmm1, %rax
-; VLNODQ-NEXT: vcvtsi2ss %rax, %xmm15, %xmm2
-; VLNODQ-NEXT: vmovq %xmm1, %rax
-; VLNODQ-NEXT: vcvtsi2ss %rax, %xmm15, %xmm1
-; VLNODQ-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[2,3]
-; VLNODQ-NEXT: vextracti32x4 $3, %zmm0, %xmm2
-; VLNODQ-NEXT: vmovq %xmm2, %rax
-; VLNODQ-NEXT: vcvtsi2ss %rax, %xmm15, %xmm3
-; VLNODQ-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1],xmm3[0],xmm1[3]
-; VLNODQ-NEXT: vpextrq $1, %xmm2, %rax
-; VLNODQ-NEXT: vcvtsi2ss %rax, %xmm15, %xmm2
-; VLNODQ-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1,2],xmm2[0]
-; VLNODQ-NEXT: vpextrq $1, %xmm0, %rax
-; VLNODQ-NEXT: vcvtsi2ss %rax, %xmm15, %xmm2
-; VLNODQ-NEXT: vmovq %xmm0, %rax
-; VLNODQ-NEXT: vcvtsi2ss %rax, %xmm15, %xmm3
-; VLNODQ-NEXT: vinsertps {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[2,3]
-; VLNODQ-NEXT: vextracti128 $1, %ymm0, %xmm0
-; VLNODQ-NEXT: vmovq %xmm0, %rax
-; VLNODQ-NEXT: vcvtsi2ss %rax, %xmm15, %xmm3
-; VLNODQ-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1],xmm3[0],xmm2[3]
-; VLNODQ-NEXT: vpextrq $1, %xmm0, %rax
-; VLNODQ-NEXT: vcvtsi2ss %rax, %xmm15, %xmm0
-; VLNODQ-NEXT: vinsertps {{.*#+}} xmm0 = xmm2[0,1,2],xmm0[0]
-; VLNODQ-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; VLNODQ-NEXT: vpmovqd %zmm0, %ymm0
+; VLNODQ-NEXT: vcvtdq2ps %ymm0, %ymm0
; VLNODQ-NEXT: retq
;
; VLDQ-LABEL: sitofp_trunc_nsw_v8i64_v8f32:
@@ -276,35 +228,17 @@ define <8 x float> @sitofp_trunc_nsw_v8i64_v8f32(<8 x i64> %a0) {
define <4 x float> @uitofp_trunc_nuw_v4i64_v4f32(<4 x i64> %a0) {
; NOVL-LABEL: uitofp_trunc_nuw_v4i64_v4f32:
; NOVL: # %bb.0:
-; NOVL-NEXT: vpextrq $1, %xmm0, %rax
-; NOVL-NEXT: vcvtusi2ss %rax, %xmm15, %xmm1
-; NOVL-NEXT: vmovq %xmm0, %rax
-; NOVL-NEXT: vcvtusi2ss %rax, %xmm15, %xmm2
-; NOVL-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[2,3]
-; NOVL-NEXT: vextracti128 $1, %ymm0, %xmm0
-; NOVL-NEXT: vmovq %xmm0, %rax
-; NOVL-NEXT: vcvtusi2ss %rax, %xmm15, %xmm2
-; NOVL-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1],xmm2[0],xmm1[3]
-; NOVL-NEXT: vpextrq $1, %xmm0, %rax
-; NOVL-NEXT: vcvtusi2ss %rax, %xmm15, %xmm0
-; NOVL-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
+; NOVL-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
+; NOVL-NEXT: vpmovqd %zmm0, %ymm0
+; NOVL-NEXT: vcvtudq2ps %zmm0, %zmm0
+; NOVL-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; NOVL-NEXT: vzeroupper
; NOVL-NEXT: retq
;
; VLNODQ-LABEL: uitofp_trunc_nuw_v4i64_v4f32:
; VLNODQ: # %bb.0:
-; VLNODQ-NEXT: vpextrq $1, %xmm0, %rax
-; VLNODQ-NEXT: vcvtusi2ss %rax, %xmm15, %xmm1
-; VLNODQ-NEXT: vmovq %xmm0, %rax
-; VLNODQ-NEXT: vcvtusi2ss %rax, %xmm15, %xmm2
-; VLNODQ-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[2,3]
-; VLNODQ-NEXT: vextracti128 $1, %ymm0, %xmm0
-; VLNODQ-NEXT: vmovq %xmm0, %rax
-; VLNODQ-NEXT: vcvtusi2ss %rax, %xmm15, %xmm2
-; VLNODQ-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1],xmm2[0],xmm1[3]
-; VLNODQ-NEXT: vpextrq $1, %xmm0, %rax
-; VLNODQ-NEXT: vcvtusi2ss %rax, %xmm15, %xmm0
-; VLNODQ-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
+; VLNODQ-NEXT: vpmovqd %ymm0, %xmm0
+; VLNODQ-NEXT: vcvtudq2ps %xmm0, %xmm0
; VLNODQ-NEXT: vzeroupper
; VLNODQ-NEXT: retq
;
diff --git a/llvm/test/CodeGen/X86/combine-sitofp.ll b/llvm/test/CodeGen/X86/combine-sitofp.ll
index 97f71f2e3a6ff0..395d048dfd1f35 100644
--- a/llvm/test/CodeGen/X86/combine-sitofp.ll
+++ b/llvm/test/CodeGen/X86/combine-sitofp.ll
@@ -1,6 +1,6 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64 | FileCheck %s --check-prefixes=SSE,SSE2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v2 | FileCheck %s --check-prefixes=SSE,SSE4
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64 | FileCheck %s --check-prefixes=SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v2 | FileCheck %s --check-prefixes=SSE
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=sandybridge | FileCheck %s --check-prefixes=AVX,AVX1OR2,AVX1
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v3 | FileCheck %s --check-prefixes=AVX,AVX1OR2,AVX2
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v4 | FileCheck %s --check-prefixes=AVX,AVX512
@@ -301,35 +301,16 @@ define <16 x float> @concat_sitofp_v16f32_v8i32(<8 x i32> %a0, <8 x i32> %a1, <8
}
define <2 x double> @sitofp_trunc_nsw_v2i64_v2f64(<2 x i64> %a0) {
-; SSE2-LABEL: sitofp_trunc_nsw_v2i64_v2f64:
-; SSE2: # %bb.0:
-; SSE2-NEXT: movq %xmm0, %rax
-; SSE2-NEXT: cvtsi2sd %rax, %xmm1
-; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
-; SSE2-NEXT: movq %xmm0, %rax
-; SSE2-NEXT: xorps %xmm0, %xmm0
-; SSE2-NEXT: cvtsi2sd %rax, %xmm0
-; SSE2-NEXT: unpcklpd {{.*#+}} xmm1 = xmm1[0],xmm0[0]
-; SSE2-NEXT: movapd %xmm1, %xmm0
-; SSE2-NEXT: retq
-;
-; SSE4-LABEL: sitofp_trunc_nsw_v2i64_v2f64:
-; SSE4: # %bb.0:
-; SSE4-NEXT: pextrq $1, %xmm0, %rax
-; SSE4-NEXT: cvtsi2sd %rax, %xmm1
-; SSE4-NEXT: movq %xmm0, %rax
-; SSE4-NEXT: xorps %xmm0, %xmm0
-; SSE4-NEXT: cvtsi2sd %rax, %xmm0
-; SSE4-NEXT: unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; SSE4-NEXT: retq
+; SSE-LABEL: sitofp_trunc_nsw_v2i64_v2f64:
+; SSE: # %bb.0:
+; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; SSE-NEXT: cvtdq2pd %xmm0, %xmm0
+; SSE-NEXT: retq
;
; AVX1OR2-LABEL: sitofp_trunc_nsw_v2i64_v2f64:
; AVX1OR2: # %bb.0:
-; AVX1OR2-NEXT: vpextrq $1, %xmm0, %rax
-; AVX1OR2-NEXT: vcvtsi2sd %rax, %xmm15, %xmm1
-; AVX1OR2-NEXT: vmovq %xmm0, %rax
-; AVX1OR2-NEXT: vcvtsi2sd %rax, %xmm15, %xmm0
-; AVX1OR2-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX1OR2-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; AVX1OR2-NEXT: vcvtdq2pd %xmm0, %xmm0
; AVX1OR2-NEXT: retq
;
; AVX512-LABEL: sitofp_trunc_nsw_v2i64_v2f64:
@@ -339,11 +320,8 @@ define <2 x double> @sitofp_trunc_nsw_v2i64_v2f64(<2 x i64> %a0) {
;
; AVX512F-LABEL: sitofp_trunc_nsw_v2i64_v2f64:
; AVX512F: # %bb.0:
-; AVX512F-NEXT: vpextrq $1, %xmm0, %rax
-; AVX512F-NEXT: vcvtsi2sd %rax, %xmm15, %xmm1
-; AVX512F-NEXT: vmovq %xmm0, %rax
-; AVX512F-NEXT: vcvtsi2sd %rax, %xmm15, %xmm0
-; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX512F-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; AVX512F-NEXT: vcvtdq2pd %xmm0, %xmm0
; AVX512F-NEXT: retq
%t = trunc nsw <2 x i64> %a0 to <2 x i32>
%r = sitofp <2 x i32> %t to <2 x double>
@@ -362,33 +340,16 @@ define <4 x double> @sitofp_trunc_nsw_v4i64_v4f64(<4 x i64> %a0) {
; AVX1-LABEL: sitofp_trunc_nsw_v4i64_v4f64:
; AVX1: # %bb.0:
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT: vpextrq $1, %xmm1, %rax
-; AVX1-NEXT: vcvtsi2sd %rax, %xmm15, %xmm2
-; AVX1-NEXT: vmovq %xmm1, %rax
-; AVX1-NEXT: vcvtsi2sd %rax, %xmm15, %xmm1
-; AVX1-NEXT: vpextrq $1, %xmm0, %rax
-; AVX1-NEXT: vcvtsi2sd %rax, %xmm15, %xmm3
-; AVX1-NEXT: vmovq %xmm0, %rax
-; AVX1-NEXT: vcvtsi2sd %rax, %xmm15, %xmm0
-; AVX1-NEXT: vunpcklpd {{.*#+}} xmm1 = xmm1[0],xmm2[0]
-; AVX1-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm3[0]
-; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX1-NEXT: vcvtdq2pd %xmm0, %ymm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: sitofp_trunc_nsw_v4i64_v4f64:
; AVX2: # %bb.0:
-; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT: vpextrq $1, %xmm1, %rax
-; AVX2-NEXT: vcvtsi2sd %rax, %xmm15, %xmm2
-; AVX2-NEXT: vmovq %xmm1, %rax
-; AVX2-NEXT: vcvtsi2sd %rax, %xmm15, %xmm1
-; AVX2-NEXT: vpextrq $1, %xmm0, %rax
-; AVX2-NEXT: vunpcklpd {{.*#+}} xmm1 = xmm1[0],xmm2[0]
-; AVX2-NEXT: vcvtsi2sd %rax, %xmm15, %xmm2
-; AVX2-NEXT: vmovq %xmm0, %rax
-; AVX2-NEXT: vcvtsi2sd %rax, %xmm15, %xmm0
-; AVX2-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm2[0]
-; AVX2-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-NEXT: vbroadcastf128 {{.*#+}} ymm1 = [0,2,4,6,0,2,4,6]
+; AVX2-NEXT: # ymm1 = mem[0,1,0,1]
+; AVX2-NEXT: vpermps %ymm0, %ymm1, %ymm0
+; AVX2-NEXT: vcvtdq2pd %xmm0, %ymm0
; AVX2-NEXT: retq
;
; AVX512-LABEL: sitofp_trunc_nsw_v4i64_v4f64:
@@ -398,18 +359,9 @@ define <4 x double> @sitofp_trunc_nsw_v4i64_v4f64(<4 x i64> %a0) {
;
; AVX512F-LABEL: sitofp_trunc_nsw_v4i64_v4f64:
; AVX512F: # %bb.0:
-; AVX512F-NEXT: vextracti128 $1, %ymm0, %xmm1
-; AVX512F-NEXT: vpextrq $1, %xmm1, %rax
-; AVX512F-NEXT: vcvtsi2sd %rax, %xmm15, %xmm2
-; AVX512F-NEXT: vmovq %xmm1, %rax
-; AVX512F-NEXT: vcvtsi2sd %rax, %xmm15, %xmm1
-; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm1 = xmm1[0],xmm2[0]
-; AVX512F-NEXT: vpextrq $1, %xmm0, %rax
-; AVX512F-NEXT: vcvtsi2sd %rax, %xmm15, %xmm2
-; AVX512F-NEXT: vmovq %xmm0, %rax
-; AVX512F-NEXT: vcvtsi2sd %rax, %xmm15, %xmm0
-; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm2[0]
-; AVX512F-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-NEXT: vpmovqd %zmm0, %ymm0
+; AVX512F-NEXT: vcvtdq2pd %xmm0, %ymm0
; AVX512F-NEXT: retq
%t = trunc nsw <4 x i64> %a0 to <4 x i32>
%r = sitofp <4 x i32> %t to <4 x double>
@@ -425,35 +377,18 @@ define <4 x float> @sitofp_trunc_nsw_v4i64_v4f32(<4 x i64> %a0) {
;
; AVX1-LABEL: sitofp_trunc_nsw_v4i64_v4f32:
; AVX1: # %bb.0:
-; AVX1-NEXT: vpextrq $1, %xmm0, %rax
-; AVX1-NEXT: vcvtsi2ss %rax, %xmm15, %xmm1
-; AVX1-NEXT: vmovq %xmm0, %rax
-; AVX1-NEXT: vcvtsi2ss %rax, %xmm15, %xmm2
-; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT: vmovq %xmm0, %rax
-; AVX1-NEXT: vcvtsi2ss %rax, %xmm15, %xmm3
-; AVX1-NEXT: vpextrq $1, %xmm0, %rax
-; AVX1-NEXT: vcvtsi2ss %rax, %xmm15, %xmm0
-; AVX1-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[2,3]
-; AVX1-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1],xmm3[0],xmm1[3]
-; AVX1-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
+; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
+; AVX1-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX1-NEXT: vcvtdq2ps %xmm0, %xmm0
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
;
; AVX2-LABEL: sitofp_trunc_nsw_v4i64_v4f32:
; AVX2: # %bb.0:
-; AVX2-NEXT: vpextrq $1, %xmm0, %rax
-; AVX2-NEXT: vcvtsi2ss %rax, %xmm15, %xmm1
-; AVX2-NEXT: vmovq %xmm0, %rax
-; AVX2-NEXT: vcvtsi2ss %rax, %xmm15, %xmm2
-; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm0
-; AVX2-NEXT: vmovq %xmm0, %rax
-; AVX2-NEXT: vcvtsi2ss %rax, %xmm15, %xmm3
-; AVX2-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[2,3]
-; AVX2-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1],xmm3[0],xmm1[3]
-; AVX2-NEXT: vpextrq $1, %xmm0, %rax
-; AVX2-NEXT: vcvtsi2ss %rax, %xmm15, %xmm0
-; AVX2-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
+; AVX2-NEXT: vbroadcastf128 {{.*#+}} ymm1 = [0,2,4,6,0,2,4,6]
+; AVX2-NEXT: # ymm1 = mem[0,1,0,1]
+; AVX2-NEXT: vpermps %ymm0, %ymm1, %ymm0
+; AVX2-NEXT: vcvtdq2ps %xmm0, %xmm0
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
@@ -465,18 +400,9 @@ define <4 x float> @sitofp_trunc_nsw_v4i64_v4f32(<4 x i64> %a0) {
;
; AVX512F-LABEL: sitofp_trunc_nsw_v4i64_v4f32:
; AVX512F: # %bb.0:
-; AVX512F-NEXT: vpextrq $1, %xmm0, %rax
-; AVX512F-NEXT: vcvtsi2ss %rax, %xmm15, %xmm1
-; AVX512F-NEXT: vmovq %xmm0, %rax
-; AVX512F-NEXT: vcvtsi2ss %rax, %xmm15, %xmm2
-; AVX512F-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[2,3]
-; AVX512F-NEXT: vextracti128 $1, %ymm0, %xmm0
-; AVX512F-NEXT: vmovq %xmm0, %rax
-; AVX512F-NEXT: vcvtsi2ss %rax, %xmm15, %xmm2
-; AVX512F-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1],xmm2[0],xmm1[3]
-; AVX512F-NEXT: vpextrq $1, %xmm0, %rax
-; AVX512F-NEXT: vcvtsi2ss %rax, %xmm15, %xmm0
-; AVX512F-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
+; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-NEXT: vpmovqd %zmm0, %ymm0
+; AVX512F-NEXT: vcvtdq2ps %xmm0, %xmm0
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
%t = trunc nsw <4 x i64> %a0 to <4 x i32>
@@ -523,32 +449,8 @@ define <8 x double> @sitofp_trunc_nsw_v8i64_v8f64(<8 x i64> %a0) {
;
; AVX512F-LABEL: sitofp_trunc_nsw_v8i64_v8f64:
; AVX512F: # %bb.0:
-; AVX512F-NEXT: vextracti32x4 $3, %zmm0, %xmm1
-; AVX512F-NEXT: vpextrq $1, %xmm1, %rax
-; AVX512F-NEXT: vcvtsi2sd %rax, %xmm15, %xmm2
-; AVX512F-NEXT: vmovq %xmm1, %rax
-; AVX512F-NEXT: vcvtsi2sd %rax, %xmm15, %xmm1
-; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm1 = xmm1[0],xmm2[0]
-; AVX512F-NEXT: vextracti32x4 $2, %zmm0, %xmm2
-; AVX512F-NEXT: vpextrq $1, %xmm2, %rax
-; AVX512F-NEXT: vcvtsi2sd %rax, %xmm15, %xmm3
-; AVX512F-NEXT: vmovq %xmm2, %rax
-; AVX512F-NEXT: vcvtsi2sd %rax, %xmm15, %xmm2
-; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
-; AVX512F-NEXT: vinsertf128 $1, %xmm1, %ymm2, %ymm1
-; AVX512F-NEXT: vextracti128 $1, %ymm0, %xmm2
-; AVX512F-NEXT: vpextrq $1, %xmm2, %rax
-; AVX512F-NEXT: vcvtsi2sd %rax, %xmm15, %xmm3
-; AVX512F-NEXT: vmovq %xmm2, %rax
-; AVX512F-NEXT: vcvtsi2sd %rax, %xmm15, %xmm2
-; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
-; AVX512F-NEXT: vpextrq $1, %xmm0, %rax
-; AVX512F-NEXT: vcvtsi2sd %rax, %xmm15, %xmm3
-; AVX512F-NEXT: vmovq %xmm0, %rax
-; AVX512F-NEXT: vcvtsi2sd %rax, %xmm15, %xmm0
-; AVX512F-NEXT: vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm3[0]
-; AVX512F-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX512F-NEXT: vinsertf64x4 $1, %ymm1, %zmm0, %zmm0
+; AVX512F-NEXT: vpmovqd %zmm0, %ymm0
+; AVX512F-NEXT: vcvtdq2pd %ymm0, %zmm0
; AVX512F-NEXT: retq
%t = trunc nsw <8 x i64> %a0 to <8 x i32>
%r = sitofp <8 x i32> %t to <8 x double>
diff --git a/llvm/test/CodeGen/X86/combine-uitofp.ll b/llvm/test/CodeGen/X86/combine-uitofp.ll
index 2d556e90499535..34c6176a90fb3f 100644
--- a/llvm/test/CodeGen/X86/combine-uitofp.ll
+++ b/llvm/test/CodeGen/X86/combine-uitofp.ll
@@ -937,46 +937,39 @@ define float @uitofp_trunc_i64_i32(i64 %a0) {
define <2 x double> @uitofp_trunc_nuw_v2i64_v2f64(<2 x i64> %a0) {
; SSE2-LABEL: uitofp_trunc_nuw_v2i64_v2f64:
; SSE2: # %bb.0:
-; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [4294967295,4294967295]
-; SSE2-NEXT: pand %xmm0, %xmm1
-; SSE2-NEXT: por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; SSE2-NEXT: psrlq $32, %xmm0
-; SSE2-NEXT: por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; SSE2-NEXT: subpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; SSE2-NEXT: addpd %xmm1, %xmm0
+; SSE2-NEXT: andpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE2-NEXT: movapd {{.*#+}} xmm1 = [4.503599627370496E+15,4.503599627370496E+15]
+; SSE2-NEXT: orpd %xmm1, %xmm0
+; SSE2-NEXT: subpd %xmm1, %xmm0
; SSE2-NEXT: retq
;
; SSE4-LABEL: uitofp_trunc_nuw_v2i64_v2f64:
; SSE4: # %bb.0:
-; SSE4-NEXT: pxor %xmm1, %xmm1
-; SSE4-NEXT: pblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
-; SSE4-NEXT: por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; SSE4-NEXT: psrlq $32, %xmm0
-; SSE4-NEXT: por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; SSE4-NEXT: subpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; SSE4-NEXT: addpd %xmm1, %xmm0
+; SSE4-NEXT: xorps %xmm1, %xmm1
+; SSE4-NEXT: blendps {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
+; SSE4-NEXT: movaps {{.*#+}} xmm1 = [4.503599627370496E+15,4.503599627370496E+15]
+; SSE4-NEXT: orps %xmm1, %xmm0
+; SSE4-NEXT: subpd %xmm1, %xmm0
; SSE4-NEXT: retq
;
; AVX1-LABEL: uitofp_trunc_nuw_v2i64_v2f64:
; AVX1: # %bb.0:
-; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
-; AVX1-NEXT: vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
-; AVX1-NEXT: vpsrlq $32, %xmm0, %xmm0
-; AVX1-NEXT: vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; AVX1-NEXT: vsubpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; AVX1-NEXT: vaddpd %xmm0, %xmm1, %xmm0
+; AVX1-NEXT: vxorps %xmm1, %xmm1, %xmm1
+; AVX1-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
+; AVX1-NEXT: vmovddup {{.*#+}} xmm1 = [4.503599627370496E+15,4.503599627370496E+15]
+; AVX1-NEXT: # xmm1 = mem[0,0]
+; AVX1-NEXT: vorps %xmm1, %xmm0, %xmm0
+; AVX1-NEXT: vsubpd %xmm1, %xmm0, %xmm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: uitofp_trunc_nuw_v2i64_v2f64:
; AVX2: # %bb.0:
-; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT: vpblendd {{.*#+}} xmm1 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
-; AVX2-NEXT: vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
-; AVX2-NEXT: vpsrlq $32, %xmm0, %xmm0
-; AVX2-NEXT: vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; AVX2-NEXT: vsubpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; AVX2-NEXT: vaddpd %xmm0, %xmm1, %xmm0
+; AVX2-NEXT: vxorps %xmm1, %xmm1, %xmm1
+; AVX2-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
+; AVX2-NEXT: vmovddup {{.*#+}} xmm1 = [4.503599627370496E+15,4.503599627370496E+15]
+; AVX2-NEXT: # xmm1 = mem[0,0]
+; AVX2-NEXT: vorps %xmm1, %xmm0, %xmm0
+; AVX2-NEXT: vsubpd %xmm1, %xmm0, %xmm0
; AVX2-NEXT: retq
;
; AVX512-LABEL: uitofp_trunc_nuw_v2i64_v2f64:
@@ -986,13 +979,10 @@ define <2 x double> @uitofp_trunc_nuw_v2i64_v2f64(<2 x i64> %a0) {
;
; AVX512F-LABEL: uitofp_trunc_nuw_v2i64_v2f64:
; AVX512F: # %bb.0:
-; AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512F-NEXT: vpblendd {{.*#+}} xmm1 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
-; AVX512F-NEXT: vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
-; AVX512F-NEXT: vpsrlq $32, %xmm0, %xmm0
-; AVX512F-NEXT: vpor {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; AVX512F-NEXT: vsubpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; AVX512F-NEXT: vaddpd %xmm0, %xmm1, %xmm0
+; AVX512F-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; AVX512F-NEXT: vcvtudq2pd %ymm0, %zmm0
+; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
+; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
%t = trunc nuw <2 x i64> %a0 to <2 x i32>
%r = uitofp <2 x i32> %t to <2 x double>
@@ -1027,27 +1017,21 @@ define <4 x double> @uitofp_trunc_nuw_v4i64_v4f64(<4 x i64> %a0) {
; AVX1-LABEL: uitofp_trunc_nuw_v4i64_v4f64:
; AVX1: # %bb.0:
; AVX1-NEXT: vxorps %xmm1, %xmm1, %xmm1
-; AVX1-NEXT: vblendps {{.*#+}} ymm2 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
-; AVX1-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
-; AVX1-NEXT: vshufps {{.*#+}} ymm0 = ymm0[1,3],ymm1[1,3],ymm0[5,7],ymm1[5,7]
-; AVX1-NEXT: vshufps {{.*#+}} ymm0 = ymm0[0,2,1,3,4,6,5,7]
-; AVX1-NEXT: vorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
-; AVX1-NEXT: vsubpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
-; AVX1-NEXT: vaddpd %ymm0, %ymm2, %ymm0
+; AVX1-NEXT: vbroadcastsd {{.*#+}} ymm2 = [4.503599627370496E+15,4.503599627370496E+15,4.503599627370496E+15,4.503599627370496E+15]
+; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
+; AVX1-NEXT: vorps %ymm2, %ymm0, %ymm0
+; AVX1-NEXT: vsubpd %ymm2, %ymm0, %ymm0
; AVX1-NEXT: retq
;
; AVX2-LABEL: uitofp_trunc_nuw_v4i64_v4f64:
; AVX2: # %bb.0:
-; AVX2-NEXT: vpsrlq $32, %ymm0, %ymm1
-; AVX2-NEXT: vpbroadcastq {{.*#+}} ymm2 = [4985484787499139072,4985484787499139072,4985484787499139072,4985484787499139072]
-; AVX2-NEXT: vpor %ymm2, %ymm1, %ymm1
-; AVX2-NEXT: vbroadcastsd {{.*#+}} ymm2 = [1.9342813118337666E+25,1.9342813118337666E+25,1.9342813118337666E+25,1.9342813118337666E+25]
-; AVX2-NEXT: vsubpd %ymm2, %ymm1, %ymm1
-; AVX2-NEXT: vxorpd %xmm2, %xmm2, %xmm2
-; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0],ymm2[1],ymm0[2],ymm2[3],ymm0[4],ymm2[5],ymm0[6],ymm2[7]
-; AVX2-NEXT: vpbroadcastq {{.*#+}} ymm2 = [4841369599423283200,4841369599423283200,4841369599423283200,4841369599423283200]
-; AVX2-NEXT: vpor %ymm2, %ymm0, %ymm0
-; AVX2-NEXT: vaddpd %ymm1, %ymm0, %ymm0
+; AVX2-NEXT: vbroadcasti128 {{.*#+}} ymm1 = [0,2,4,6,0,2,4,6]
+; AVX2-NEXT: # ymm1 = mem[0,1,0,1]
+; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
+; AVX2-NEXT: vpmovzxdq {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
+; AVX2-NEXT: vpbroadcastq {{.*#+}} ymm1 = [4.503599627370496E+15,4.503599627370496E+15,4.503599627370496E+15,4.503599627370496E+15]
+; AVX2-NEXT: vpor %ymm1, %ymm0, %ymm0
+; AVX2-NEXT: vsubpd %ymm1, %ymm0, %ymm0
; AVX2-NEXT: retq
;
; AVX512-LABEL: uitofp_trunc_nuw_v4i64_v4f64:
@@ -1057,16 +1041,10 @@ define <4 x double> @uitofp_trunc_nuw_v4i64_v4f64(<4 x i64> %a0) {
;
; AVX512F-LABEL: uitofp_trunc_nuw_v4i64_v4f64:
; AVX512F: # %bb.0:
-; AVX512F-NEXT: vpsrlq $32, %ymm0, %ymm1
-; AVX512F-NEXT: vpbroadcastq {{.*#+}} ymm2 = [4985484787499139072,4985484787499139072,4985484787499139072,4985484787499139072]
-; AVX512F-NEXT: vpor %ymm2, %ymm1, %ymm1
-; AVX512F-NEXT: vbroadcastsd {{.*#+}} ymm2 = [1.9342813118337666E+25,1.9342813118337666E+25,1.9342813118337666E+25,1.9342813118337666E+25]
-; AVX512F-NEXT: vsubpd %ymm2, %ymm1, %ymm1
-; AVX512F-NEXT: vxorpd %xmm2, %xmm2, %xmm2
-; AVX512F-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0],ymm2[1],ymm0[2],ymm2[3],ymm0[4],ymm2[5],ymm0[6],ymm2[7]
-; AVX512F-NEXT: vpbroadcastq {{.*#+}} ymm2 = [4841369599423283200,4841369599423283200,4841369599423283200,4841369599423283200]
-; AVX512F-NEXT: vpor %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT: vaddpd %ymm1, %ymm0, %ymm0
+; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-NEXT: vpmovqd %zmm0, %ymm0
+; AVX512F-NEXT: vcvtudq2pd %ymm0, %zmm0
+; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 killed $zmm0
; AVX512F-NEXT: retq
%t = trunc nuw <4 x i64> %a0 to <4 x i32>
%r = uitofp <4 x i32> %t to <4 x double>
@@ -1100,59 +1078,28 @@ define <4 x float> @uitofp_trunc_nuw_v4i64_v4f32(<4 x i64> %a0) {
; AVX1-LABEL: uitofp_trunc_nuw_v4i64_v4f32:
; AVX1: # %bb.0:
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; AVX1-NEXT: vpcmpgtq %xmm1, %xmm2, %xmm3
-; AVX1-NEXT: vpsrlq $1, %xmm0, %xmm4
-; AVX1-NEXT: vpsrlq $1, %xmm1, %xmm5
-; AVX1-NEXT: vinsertf128 $1, %xmm5, %ymm4, %ymm4
-; AVX1-NEXT: vandpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm5
-; AVX1-NEXT: vorpd %ymm5, %ymm4, %ymm4
-; AVX1-NEXT: vblendvpd %xmm0, %xmm4, %xmm0, %xmm5
-; AVX1-NEXT: vpextrq $1, %xmm5, %rax
-; AVX1-NEXT: vpcmpgtq %xmm0, %xmm2, %xmm0
-; AVX1-NEXT: vcvtsi2ss %rax, %xmm15, %xmm2
-; AVX1-NEXT: vmovq %xmm5, %rax
-; AVX1-NEXT: vcvtsi2ss %rax, %xmm15, %xmm5
-; AVX1-NEXT: vpackssdw %xmm3, %xmm0, %xmm0
-; AVX1-NEXT: vextractf128 $1, %ymm4, %xmm3
-; AVX1-NEXT: vblendvpd %xmm1, %xmm3, %xmm1, %xmm1
-; AVX1-NEXT: vmovq %xmm1, %rax
-; AVX1-NEXT: vcvtsi2ss %rax, %xmm15, %xmm3
-; AVX1-NEXT: vinsertps {{.*#+}} xmm2 = xmm5[0],xmm2[0],xmm5[2,3]
-; AVX1-NEXT: vinsertps {{.*#+}} xmm2 = xmm2[0,1],xmm3[0],xmm2[3]
-; AVX1-NEXT: vpextrq $1, %xmm1, %rax
-; AVX1-NEXT: vcvtsi2ss %rax, %xmm15, %xmm1
-; AVX1-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0,1,2],xmm1[0]
-; AVX1-NEXT: vaddps %xmm1, %xmm1, %xmm2
-; AVX1-NEXT: vblendvps %xmm0, %xmm2, %xmm1, %xmm0
+; AVX1-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm0[0],mem[1],xmm0[2],mem[3],xmm0[4],mem[5],xmm0[6],mem[7]
+; AVX1-NEXT: vpsrld $16, %xmm0, %xmm0
+; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],mem[1],xmm0[2],mem[3],xmm0[4],mem[5],xmm0[6],mem[7]
+; AVX1-NEXT: vsubps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT: vaddps %xmm0, %xmm1, %xmm0
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
;
; AVX2-LABEL: uitofp_trunc_nuw_v4i64_v4f32:
; AVX2: # %bb.0:
-; AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT: vpcmpgtq %ymm0, %ymm1, %ymm1
-; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm2
-; AVX2-NEXT: vpbroadcastq {{.*#+}} ymm3 = [1,1,1,1]
-; AVX2-NEXT: vpand %ymm3, %ymm0, %ymm3
-; AVX2-NEXT: vpsrlq $1, %ymm0, %ymm4
-; AVX2-NEXT: vpor %ymm3, %ymm4, %ymm3
-; AVX2-NEXT: vblendvpd %ymm0, %ymm3, %ymm0, %ymm0
-; AVX2-NEXT: vpextrq $1, %xmm0, %rax
-; AVX2-NEXT: vcvtsi2ss %rax, %xmm15, %xmm3
-; AVX2-NEXT: vmovq %xmm0, %rax
-; AVX2-NEXT: vcvtsi2ss %rax, %xmm15, %xmm4
-; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm0
-; AVX2-NEXT: vmovq %xmm0, %rax
-; AVX2-NEXT: vcvtsi2ss %rax, %xmm15, %xmm5
-; AVX2-NEXT: vpackssdw %xmm2, %xmm1, %xmm1
-; AVX2-NEXT: vinsertps {{.*#+}} xmm2 = xmm4[0],xmm3[0],xmm4[2,3]
-; AVX2-NEXT: vpextrq $1, %xmm0, %rax
-; AVX2-NEXT: vinsertps {{.*#+}} xmm0 = xmm2[0,1],xmm5[0],xmm2[3]
-; AVX2-NEXT: vcvtsi2ss %rax, %xmm15, %xmm2
-; AVX2-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[0,1,2],xmm2[0]
-; AVX2-NEXT: vaddps %xmm0, %xmm0, %xmm2
-; AVX2-NEXT: vblendvps %xmm1, %xmm2, %xmm0, %xmm0
+; AVX2-NEXT: vbroadcasti128 {{.*#+}} ymm1 = [0,2,4,6,0,2,4,6]
+; AVX2-NEXT: # ymm1 = mem[0,1,0,1]
+; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0
+; AVX2-NEXT: vpbroadcastd {{.*#+}} xmm1 = [1258291200,1258291200,1258291200,1258291200]
+; AVX2-NEXT: vpblendw {{.*#+}} xmm1 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7]
+; AVX2-NEXT: vpsrld $16, %xmm0, %xmm0
+; AVX2-NEXT: vpbroadcastd {{.*#+}} xmm2 = [1392508928,1392508928,1392508928,1392508928]
+; AVX2-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7]
+; AVX2-NEXT: vbroadcastss {{.*#+}} xmm2 = [5.49764202E+11,5.49764202E+11,5.49764202E+11,5.49764202E+11]
+; AVX2-NEXT: vsubps %xmm2, %xmm0, %xmm0
+; AVX2-NEXT: vaddps %xmm0, %xmm1, %xmm0
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
;
@@ -1164,18 +1111,10 @@ define <4 x float> @uitofp_trunc_nuw_v4i64_v4f32(<4 x i64> %a0) {
;
; AVX512F-LABEL: uitofp_trunc_nuw_v4i64_v4f32:
; AVX512F: # %bb.0:
-; AVX512F-NEXT: vpextrq $1, %xmm0, %rax
-; AVX512F-NEXT: vcvtusi2ss %rax, %xmm15, %xmm1
-; AVX512F-NEXT: vmovq %xmm0, %rax
-; AVX512F-NEXT: vcvtusi2ss %rax, %xmm15, %xmm2
-; AVX512F-NEXT: vinsertps {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[2,3]
-; AVX512F-NEXT: vextracti128 $1, %ymm0, %xmm0
-; AVX512F-NEXT: vmovq %xmm0, %rax
-; AVX512F-NEXT: vcvtusi2ss %rax, %xmm15, %xmm2
-; AVX512F-NEXT: vinsertps {{.*#+}} xmm1 = xmm1[0,1],xmm2[0],xmm1[3]
-; AVX512F-NEXT: vpextrq $1, %xmm0, %rax
-; AVX512F-NEXT: vcvtusi2ss %rax, %xmm15, %xmm0
-; AVX512F-NEXT: vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
+; AVX512F-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-NEXT: vpmovqd %zmm0, %ymm0
+; AVX512F-NEXT: vcvtudq2ps %zmm0, %zmm0
+; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512F-NEXT: vzeroupper
; AVX512F-NEXT: retq
%t = trunc nuw <4 x i64> %a0 to <4 x i32>
@@ -1253,12 +1192,8 @@ define <8 x double> @uitofp_trunc_nuw_v8i64_v8f64(<8 x i64> %a0) {
;
; AVX512F-LABEL: uitofp_trunc_nuw_v8i64_v8f64:
; AVX512F: # %bb.0:
-; AVX512F-NEXT: vpbroadcastq {{.*#+}} zmm1 = [4841369599423283200,4841369599423283200,4841369599423283200,4841369599423283200,4841369599423283200,4841369599423283200,4841369599423283200,4841369599423283200]
-; AVX512F-NEXT: vpternlogq {{.*#+}} zmm1 = zmm1 | (zmm0 & m64bcst)
-; AVX512F-NEXT: vpsrlq $32, %zmm0, %zmm0
-; AVX512F-NEXT: vporq {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm0
-; AVX512F-NEXT: vsubpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm0
-; AVX512F-NEXT: vaddpd %zmm0, %zmm1, %zmm0
+; AVX512F-NEXT: vpmovqd %zmm0, %ymm0
+; AVX512F-NEXT: vcvtudq2pd %ymm0, %zmm0
; AVX512F-NEXT: retq
%t = trunc nuw <8 x i64> %a0 to <8 x i32>
%r = uitofp <8 x i32> %t to <8 x double>
More information about the llvm-branch-commits
mailing list