[llvm] [X86] Faster truncf and roundf on x86 SSE2 (PR #226513)
Divyansh Yadav via llvm-commits
llvm-commits at lists.llvm.org
Fri Sep 25 08:55:50 PDT 2026
https://github.com/schizophrenicmaniac updated https://github.com/llvm/llvm-project/pull/226513
>From f5e4ee2aa98a781e04dea76d10828c387abdc2ca Mon Sep 17 00:00:00 2001
From: Divyansh <anshmcs at gmail.com>
Date: Fri, 25 Sep 2026 20:20:38 +0530
Subject: [PATCH 1/2] [X86] Faster truncf and roundf on x86 SSE2
---
llvm/lib/Target/X86/X86ISelLowering.cpp | 68 ++-
llvm/test/Analysis/CostModel/X86/fround.ll | 18 +-
llvm/test/CodeGen/X86/fp-round.ll | 539 ++++++++++--------
llvm/test/CodeGen/X86/fp16-libcalls.ll | 35 +-
llvm/test/CodeGen/X86/freeze-unary.ll | 41 +-
llvm/test/CodeGen/X86/ftrunc.ll | 359 +++++++-----
llvm/test/CodeGen/X86/isel-ftrunc.ll | 59 +-
llvm/test/CodeGen/X86/isint.ll | 96 +++-
llvm/test/CodeGen/X86/llround-conv.ll | 32 +-
llvm/test/CodeGen/X86/lround-conv-i32.ll | 41 +-
llvm/test/CodeGen/X86/lround-conv-i64.ll | 16 +-
llvm/test/CodeGen/X86/unpredictable-brcond.ll | 29 +-
.../Transforms/SLPVectorizer/X86/fround.ll | 168 ++----
13 files changed, 905 insertions(+), 596 deletions(-)
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index e2f3b5f3cd3d5..c819efa95a515 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -632,6 +632,11 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::f32, Expand);
for (auto VT : { MVT::f32, MVT::f64 }) {
+ if (VT == MVT::f32 || Subtarget.is64Bit()) {
+ setOperationAction(ISD::FTRUNC, VT, Custom);
+ setOperationAction(ISD::FROUND, VT, Custom);
+ }
+
// Use ANDPD to simulate FABS.
setOperationAction(ISD::FABS, VT, Custom);
@@ -1096,6 +1101,13 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
for (auto VT : { MVT::f64, MVT::v4f32, MVT::v2f64 })
SetFPMinMaxAction(VT);
+ for (auto VT : { MVT::v4f32, MVT::v2f64 }) {
+ if (VT == MVT::v4f32 || Subtarget.is64Bit()) {
+ setOperationAction(ISD::FTRUNC, VT, Custom);
+ setOperationAction(ISD::FROUND, VT, Custom);
+ }
+ }
+
setOperationAction(ISD::MUL, MVT::v2i8, Custom);
setOperationAction(ISD::MUL, MVT::v4i8, Custom);
setOperationAction(ISD::MUL, MVT::v8i8, Custom);
@@ -23178,6 +23190,51 @@ SDValue X86TargetLowering::lowerFaddFsub(SDValue Op, SelectionDAG &DAG) const {
return lowerAddSubToHorizontalOp(Op, SDLoc(Op), DAG, Subtarget);
}
+static SDValue lowerFTRUNC_FROUND_SSE2(SDValue Op, SelectionDAG &DAG) {
+ SDLoc DL(Op);
+ SDValue N0 = Op.getOperand(0);
+ MVT VT = Op.getSimpleValueType();
+ bool IsRound = Op.getOpcode() == ISD::FROUND;
+
+ SDValue Abs = DAG.getNode(ISD::FABS, DL, VT, N0);
+ SDValue AbsBiased = Abs;
+ if (IsRound) {
+ const fltSemantics &Sem = VT.getFltSemantics();
+ APFloat Bias = APFloat(0.5f);
+ bool Ignored;
+ Bias.convert(Sem, APFloat::rmNearestTiesToEven, &Ignored);
+ Bias.next(/*nextDown*/true);
+ AbsBiased = DAG.getNode(ISD::FADD, DL, VT, Abs, DAG.getConstantFP(Bias, DL, VT));
+ }
+
+ MVT IntVT;
+ if (VT == MVT::f32) IntVT = MVT::i32;
+ else if (VT == MVT::f64) IntVT = MVT::i64;
+ else if (VT == MVT::v4f32) IntVT = MVT::v4i32;
+ else if (VT == MVT::v2f64) IntVT = MVT::v2i64;
+ else llvm_unreachable("Unexpected type");
+
+ SDValue AbsInt = DAG.getBitcast(IntVT, Abs);
+
+ SDValue Threshold;
+ if (VT.getScalarType() == MVT::f32) {
+ Threshold = DAG.getConstant(0x4EFFFFFF, DL, IntVT);
+ } else {
+ Threshold = DAG.getConstant(0x432FFFFFFFFFFFFFULL, DL, IntVT);
+ }
+
+ EVT CCVT = DAG.getTargetLoweringInfo().getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(), IntVT);
+ SDValue IsLarge = DAG.getSetCC(DL, CCVT, AbsInt, Threshold, ISD::SETGT);
+
+ SDValue TruncInt = DAG.getNode(ISD::FP_TO_SINT, DL, IntVT, AbsBiased);
+ SDValue AbsTrunc = DAG.getNode(ISD::SINT_TO_FP, DL, VT, TruncInt);
+
+ SDValue Trunc = DAG.getNode(ISD::FCOPYSIGN, DL, VT, AbsTrunc, N0);
+
+ unsigned SelOpc = VT.isVector() ? ISD::VSELECT : ISD::SELECT;
+ return DAG.getNode(SelOpc, DL, VT, IsLarge, N0, Trunc);
+}
+
/// ISD::FROUND is defined to round to nearest with ties rounding away from 0.
/// This mode isn't supported in hardware on X86. But as long as we aren't
/// compiling with trapping math, we can emulate this with
@@ -34849,7 +34906,16 @@ SDValue X86TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
case ISD::STORE: return LowerStore(Op, Subtarget, DAG);
case ISD::FADD:
case ISD::FSUB: return lowerFaddFsub(Op, DAG);
- case ISD::FROUND: return LowerFROUND(Op, DAG);
+ case ISD::FTRUNC:
+ case ISD::FROUND: {
+ MVT VT = Op.getSimpleValueType();
+ if (Subtarget.hasSSE2() && !Subtarget.hasSSE41() &&
+ (VT == MVT::f32 || VT == MVT::v4f32 || Subtarget.is64Bit()))
+ return lowerFTRUNC_FROUND_SSE2(Op, DAG);
+ if (Op.getOpcode() == ISD::FROUND)
+ return LowerFROUND(Op, DAG);
+ return SDValue();
+ }
case ISD::FABS:
case ISD::FNEG: return LowerFABSorFNEG(Op, DAG);
case ISD::FCOPYSIGN: return LowerFCOPYSIGN(Op, DAG);
diff --git a/llvm/test/Analysis/CostModel/X86/fround.ll b/llvm/test/Analysis/CostModel/X86/fround.ll
index ed7965af202d9..e40319462b116 100644
--- a/llvm/test/Analysis/CostModel/X86/fround.ll
+++ b/llvm/test/Analysis/CostModel/X86/fround.ll
@@ -267,15 +267,15 @@ define i32 @rint(i32 %arg) {
define i32 @trunc(i32 %arg) {
; SSE2-LABEL: 'trunc'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %F32 = call float @llvm.trunc.f32(float undef)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %V2F32 = call <2 x float> @llvm.trunc.v2f32(<2 x float> undef)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 43 for instruction: %V4F32 = call <4 x float> @llvm.trunc.v4f32(<4 x float> undef)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 86 for instruction: %V8F32 = call <8 x float> @llvm.trunc.v8f32(<8 x float> undef)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 172 for instruction: %V16F32 = call <16 x float> @llvm.trunc.v16f32(<16 x float> undef)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %F64 = call double @llvm.trunc.f64(double undef)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %V2F64 = call <2 x double> @llvm.trunc.v2f64(<2 x double> undef)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 42 for instruction: %V4F64 = call <4 x double> @llvm.trunc.v4f64(<4 x double> undef)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 84 for instruction: %V8F64 = call <8 x double> @llvm.trunc.v8f64(<8 x double> undef)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %F32 = call float @llvm.trunc.f32(float undef)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2F32 = call <2 x float> @llvm.trunc.v2f32(<2 x float> undef)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4F32 = call <4 x float> @llvm.trunc.v4f32(<4 x float> undef)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8F32 = call <8 x float> @llvm.trunc.v8f32(<8 x float> undef)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16F32 = call <16 x float> @llvm.trunc.v16f32(<16 x float> undef)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %F64 = call double @llvm.trunc.f64(double undef)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2F64 = call <2 x double> @llvm.trunc.v2f64(<2 x double> undef)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4F64 = call <4 x double> @llvm.trunc.v4f64(<4 x double> undef)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8F64 = call <8 x double> @llvm.trunc.v8f64(<8 x double> undef)
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; SSE42-LABEL: 'trunc'
diff --git a/llvm/test/CodeGen/X86/fp-round.ll b/llvm/test/CodeGen/X86/fp-round.ll
index 8595b63fc8107..145a922e12b9e 100644
--- a/llvm/test/CodeGen/X86/fp-round.ll
+++ b/llvm/test/CodeGen/X86/fp-round.ll
@@ -11,7 +11,21 @@ define half @round_f16(half %h) {
; SSE2-NEXT: pushq %rax
; SSE2-NEXT: .cfi_def_cfa_offset 16
; SSE2-NEXT: callq __extendhfsf2 at PLT
-; SSE2-NEXT: callq roundf at PLT
+; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN]
+; SSE2-NEXT: movdqa %xmm0, %xmm2
+; SSE2-NEXT: pand %xmm1, %xmm2
+; SSE2-NEXT: movd %xmm2, %eax
+; SSE2-NEXT: cmpl $1325400064, %eax # imm = 0x4F000000
+; SSE2-NEXT: jge .LBB0_2
+; SSE2-NEXT: # %bb.1: # %entry
+; SSE2-NEXT: addss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; SSE2-NEXT: cvttps2dq %xmm2, %xmm2
+; SSE2-NEXT: cvtdq2ps %xmm2, %xmm2
+; SSE2-NEXT: andps %xmm1, %xmm2
+; SSE2-NEXT: pandn %xmm0, %xmm1
+; SSE2-NEXT: por %xmm2, %xmm1
+; SSE2-NEXT: movdqa %xmm1, %xmm0
+; SSE2-NEXT: .LBB0_2: # %entry
; SSE2-NEXT: callq __truncsfhf2 at PLT
; SSE2-NEXT: popq %rax
; SSE2-NEXT: .cfi_def_cfa_offset 8
@@ -74,7 +88,22 @@ entry:
define float @round_f32(float %x) {
; SSE2-LABEL: round_f32:
; SSE2: # %bb.0:
-; SSE2-NEXT: jmp roundf at PLT # TAILCALL
+; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN]
+; SSE2-NEXT: movdqa %xmm0, %xmm2
+; SSE2-NEXT: pand %xmm1, %xmm2
+; SSE2-NEXT: movd %xmm2, %eax
+; SSE2-NEXT: cmpl $1325400064, %eax # imm = 0x4F000000
+; SSE2-NEXT: jge .LBB1_2
+; SSE2-NEXT: # %bb.1:
+; SSE2-NEXT: addss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; SSE2-NEXT: cvttps2dq %xmm2, %xmm2
+; SSE2-NEXT: cvtdq2ps %xmm2, %xmm2
+; SSE2-NEXT: andps %xmm1, %xmm2
+; SSE2-NEXT: pandn %xmm0, %xmm1
+; SSE2-NEXT: por %xmm2, %xmm1
+; SSE2-NEXT: movdqa %xmm1, %xmm0
+; SSE2-NEXT: .LBB1_2:
+; SSE2-NEXT: retq
;
; SSE41-LABEL: round_f32:
; SSE41: # %bb.0:
@@ -117,7 +146,24 @@ define float @round_f32(float %x) {
define double @round_f64(double %x) {
; SSE2-LABEL: round_f64:
; SSE2: # %bb.0:
-; SSE2-NEXT: jmp round at PLT # TAILCALL
+; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN]
+; SSE2-NEXT: movdqa %xmm0, %xmm2
+; SSE2-NEXT: pand %xmm1, %xmm2
+; SSE2-NEXT: movq %xmm2, %rax
+; SSE2-NEXT: movabsq $4841369599423283199, %rcx # imm = 0x432FFFFFFFFFFFFF
+; SSE2-NEXT: cmpq %rcx, %rax
+; SSE2-NEXT: jg .LBB2_2
+; SSE2-NEXT: # %bb.1:
+; SSE2-NEXT: addsd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; SSE2-NEXT: cvttsd2si %xmm2, %rax
+; SSE2-NEXT: xorps %xmm2, %xmm2
+; SSE2-NEXT: cvtsi2sd %rax, %xmm2
+; SSE2-NEXT: andpd %xmm1, %xmm2
+; SSE2-NEXT: pandn %xmm0, %xmm1
+; SSE2-NEXT: por %xmm2, %xmm1
+; SSE2-NEXT: movdqa %xmm1, %xmm0
+; SSE2-NEXT: .LBB2_2:
+; SSE2-NEXT: retq
;
; SSE41-LABEL: round_f64:
; SSE41: # %bb.0:
@@ -161,31 +207,20 @@ define double @round_f64(double %x) {
define <4 x float> @round_v4f32(<4 x float> %x) {
; SSE2-LABEL: round_v4f32:
; SSE2: # %bb.0:
-; SSE2-NEXT: subq $56, %rsp
-; SSE2-NEXT: .cfi_def_cfa_offset 64
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: movhlps {{.*#+}} xmm0 = xmm0[1,1]
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: unpcklps (%rsp), %xmm0 # 16-byte Folded Reload
-; SSE2-NEXT: # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; SSE2-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE2-NEXT: unpcklps {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSE2-NEXT: unpcklpd (%rsp), %xmm1 # 16-byte Folded Reload
-; SSE2-NEXT: # xmm1 = xmm1[0],mem[0]
-; SSE2-NEXT: movaps %xmm1, %xmm0
-; SSE2-NEXT: addq $56, %rsp
-; SSE2-NEXT: .cfi_def_cfa_offset 8
+; SSE2-NEXT: movaps {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN]
+; SSE2-NEXT: movaps %xmm0, %xmm2
+; SSE2-NEXT: andps %xmm1, %xmm2
+; SSE2-NEXT: movaps {{.*#+}} xmm3 = [4.9999997E-1,4.9999997E-1,4.9999997E-1,4.9999997E-1]
+; SSE2-NEXT: addps %xmm2, %xmm3
+; SSE2-NEXT: cvttps2dq %xmm3, %xmm3
+; SSE2-NEXT: cvtdq2ps %xmm3, %xmm3
+; SSE2-NEXT: andps %xmm1, %xmm3
+; SSE2-NEXT: andnps %xmm0, %xmm1
+; SSE2-NEXT: orps %xmm1, %xmm3
+; SSE2-NEXT: pcmpgtd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; SSE2-NEXT: andps %xmm2, %xmm0
+; SSE2-NEXT: andnps %xmm3, %xmm2
+; SSE2-NEXT: orps %xmm2, %xmm0
; SSE2-NEXT: retq
;
; SSE41-LABEL: round_v4f32:
@@ -227,19 +262,25 @@ define <4 x float> @round_v4f32(<4 x float> %x) {
define <2 x double> @round_v2f64(<2 x double> %x) {
; SSE2-LABEL: round_v2f64:
; SSE2: # %bb.0:
-; SSE2-NEXT: subq $40, %rsp
-; SSE2-NEXT: .cfi_def_cfa_offset 48
-; SSE2-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill
-; SSE2-NEXT: callq round at PLT
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps (%rsp), %xmm0 # 16-byte Reload
-; SSE2-NEXT: movhlps {{.*#+}} xmm0 = xmm0[1,1]
-; SSE2-NEXT: callq round at PLT
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE2-NEXT: movlhps {{.*#+}} xmm1 = xmm1[0],xmm0[0]
-; SSE2-NEXT: movaps %xmm1, %xmm0
-; SSE2-NEXT: addq $40, %rsp
-; SSE2-NEXT: .cfi_def_cfa_offset 8
+; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN]
+; SSE2-NEXT: movdqa %xmm0, %xmm2
+; SSE2-NEXT: pand %xmm1, %xmm2
+; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
+; SSE2-NEXT: addpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; SSE2-NEXT: cvttsd2si %xmm2, %rax
+; SSE2-NEXT: cvtsi2sd %rax, %xmm4
+; SSE2-NEXT: unpckhpd {{.*#+}} xmm2 = xmm2[1,1]
+; SSE2-NEXT: cvttsd2si %xmm2, %rax
+; SSE2-NEXT: xorps %xmm2, %xmm2
+; SSE2-NEXT: cvtsi2sd %rax, %xmm2
+; SSE2-NEXT: unpcklpd {{.*#+}} xmm4 = xmm4[0],xmm2[0]
+; SSE2-NEXT: andpd %xmm1, %xmm4
+; SSE2-NEXT: pandn %xmm0, %xmm1
+; SSE2-NEXT: por %xmm4, %xmm1
+; SSE2-NEXT: pcmpgtd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
+; SSE2-NEXT: pand %xmm3, %xmm0
+; SSE2-NEXT: pandn %xmm1, %xmm3
+; SSE2-NEXT: por %xmm3, %xmm0
; SSE2-NEXT: retq
;
; SSE41-LABEL: round_v2f64:
@@ -281,53 +322,35 @@ define <2 x double> @round_v2f64(<2 x double> %x) {
define <8 x float> @round_v8f32(<8 x float> %x) {
; SSE2-LABEL: round_v8f32:
; SSE2: # %bb.0:
-; SSE2-NEXT: subq $72, %rsp
-; SSE2-NEXT: .cfi_def_cfa_offset 80
-; SSE2-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill
-; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps (%rsp), %xmm0 # 16-byte Reload
-; SSE2-NEXT: movhlps {{.*#+}} xmm0 = xmm0[1,1]
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: unpcklps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; SSE2-NEXT: # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps (%rsp), %xmm0 # 16-byte Reload
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps (%rsp), %xmm0 # 16-byte Reload
-; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE2-NEXT: unpcklps {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSE2-NEXT: unpcklpd {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE2-NEXT: # xmm1 = xmm1[0],mem[0]
-; SSE2-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: movhlps {{.*#+}} xmm0 = xmm0[1,1]
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: unpcklps (%rsp), %xmm0 # 16-byte Folded Reload
-; SSE2-NEXT: # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; SSE2-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE2-NEXT: unpcklps {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSE2-NEXT: unpcklpd (%rsp), %xmm1 # 16-byte Folded Reload
-; SSE2-NEXT: # xmm1 = xmm1[0],mem[0]
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: addq $72, %rsp
-; SSE2-NEXT: .cfi_def_cfa_offset 8
+; SSE2-NEXT: movaps {{.*#+}} xmm3 = [NaN,NaN,NaN,NaN]
+; SSE2-NEXT: movaps %xmm0, %xmm2
+; SSE2-NEXT: andps %xmm3, %xmm2
+; SSE2-NEXT: movaps {{.*#+}} xmm4 = [4.9999997E-1,4.9999997E-1,4.9999997E-1,4.9999997E-1]
+; SSE2-NEXT: movaps %xmm2, %xmm5
+; SSE2-NEXT: addps %xmm4, %xmm5
+; SSE2-NEXT: cvttps2dq %xmm5, %xmm5
+; SSE2-NEXT: cvtdq2ps %xmm5, %xmm5
+; SSE2-NEXT: andps %xmm3, %xmm5
+; SSE2-NEXT: movaps %xmm3, %xmm6
+; SSE2-NEXT: movaps %xmm1, %xmm7
+; SSE2-NEXT: andps %xmm3, %xmm7
+; SSE2-NEXT: addps %xmm7, %xmm4
+; SSE2-NEXT: cvttps2dq %xmm4, %xmm4
+; SSE2-NEXT: cvtdq2ps %xmm4, %xmm4
+; SSE2-NEXT: andps %xmm3, %xmm4
+; SSE2-NEXT: andnps %xmm0, %xmm3
+; SSE2-NEXT: orps %xmm3, %xmm5
+; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [1325400063,1325400063,1325400063,1325400063]
+; SSE2-NEXT: pcmpgtd %xmm3, %xmm2
+; SSE2-NEXT: andps %xmm2, %xmm0
+; SSE2-NEXT: andnps %xmm5, %xmm2
+; SSE2-NEXT: orps %xmm2, %xmm0
+; SSE2-NEXT: andnps %xmm1, %xmm6
+; SSE2-NEXT: orps %xmm6, %xmm4
+; SSE2-NEXT: pcmpgtd %xmm3, %xmm7
+; SSE2-NEXT: andps %xmm7, %xmm1
+; SSE2-NEXT: andnps %xmm4, %xmm7
+; SSE2-NEXT: orps %xmm7, %xmm1
; SSE2-NEXT: retq
;
; SSE41-LABEL: round_v8f32:
@@ -375,29 +398,47 @@ define <8 x float> @round_v8f32(<8 x float> %x) {
define <4 x double> @round_v4f64(<4 x double> %x) {
; SSE2-LABEL: round_v4f64:
; SSE2: # %bb.0:
-; SSE2-NEXT: subq $56, %rsp
-; SSE2-NEXT: .cfi_def_cfa_offset 64
-; SSE2-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill
-; SSE2-NEXT: callq round at PLT
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps (%rsp), %xmm0 # 16-byte Reload
-; SSE2-NEXT: movhlps {{.*#+}} xmm0 = xmm0[1,1]
-; SSE2-NEXT: callq round at PLT
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE2-NEXT: movlhps {{.*#+}} xmm1 = xmm1[0],xmm0[0]
-; SSE2-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: callq round at PLT
-; SSE2-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: movhlps {{.*#+}} xmm0 = xmm0[1,1]
-; SSE2-NEXT: callq round at PLT
-; SSE2-NEXT: movaps (%rsp), %xmm1 # 16-byte Reload
-; SSE2-NEXT: movlhps {{.*#+}} xmm1 = xmm1[0],xmm0[0]
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: addq $56, %rsp
-; SSE2-NEXT: .cfi_def_cfa_offset 8
+; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [NaN,NaN]
+; SSE2-NEXT: movdqa %xmm0, %xmm4
+; SSE2-NEXT: pand %xmm2, %xmm4
+; SSE2-NEXT: movapd {{.*#+}} xmm3 = [4.9999999999999994E-1,4.9999999999999994E-1]
+; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm4[1,1,3,3]
+; SSE2-NEXT: addpd %xmm3, %xmm4
+; SSE2-NEXT: cvttsd2si %xmm4, %rax
+; SSE2-NEXT: cvtsi2sd %rax, %xmm6
+; SSE2-NEXT: unpckhpd {{.*#+}} xmm4 = xmm4[1,1]
+; SSE2-NEXT: cvttsd2si %xmm4, %rax
+; SSE2-NEXT: xorps %xmm4, %xmm4
+; SSE2-NEXT: cvtsi2sd %rax, %xmm4
+; SSE2-NEXT: unpcklpd {{.*#+}} xmm6 = xmm6[0],xmm4[0]
+; SSE2-NEXT: andpd %xmm2, %xmm6
+; SSE2-NEXT: movdqa %xmm2, %xmm4
+; SSE2-NEXT: pandn %xmm0, %xmm4
+; SSE2-NEXT: por %xmm6, %xmm4
+; SSE2-NEXT: movdqa {{.*#+}} xmm6 = [1127219199,1127219199,1127219199,1127219199]
+; SSE2-NEXT: pcmpgtd %xmm6, %xmm5
+; SSE2-NEXT: pand %xmm5, %xmm0
+; SSE2-NEXT: pandn %xmm4, %xmm5
+; SSE2-NEXT: por %xmm5, %xmm0
+; SSE2-NEXT: movapd %xmm1, %xmm4
+; SSE2-NEXT: andpd %xmm2, %xmm4
+; SSE2-NEXT: addpd %xmm4, %xmm3
+; SSE2-NEXT: cvttsd2si %xmm3, %rax
+; SSE2-NEXT: xorps %xmm5, %xmm5
+; SSE2-NEXT: cvtsi2sd %rax, %xmm5
+; SSE2-NEXT: unpckhpd {{.*#+}} xmm3 = xmm3[1,1]
+; SSE2-NEXT: cvttsd2si %xmm3, %rax
+; SSE2-NEXT: xorps %xmm3, %xmm3
+; SSE2-NEXT: cvtsi2sd %rax, %xmm3
+; SSE2-NEXT: unpcklpd {{.*#+}} xmm5 = xmm5[0],xmm3[0]
+; SSE2-NEXT: andpd %xmm2, %xmm5
+; SSE2-NEXT: andnpd %xmm1, %xmm2
+; SSE2-NEXT: orpd %xmm5, %xmm2
+; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm4[1,1,3,3]
+; SSE2-NEXT: pcmpgtd %xmm6, %xmm3
+; SSE2-NEXT: andpd %xmm3, %xmm1
+; SSE2-NEXT: andnpd %xmm2, %xmm3
+; SSE2-NEXT: orpd %xmm3, %xmm1
; SSE2-NEXT: retq
;
; SSE41-LABEL: round_v4f64:
@@ -445,99 +486,63 @@ define <4 x double> @round_v4f64(<4 x double> %x) {
define <16 x float> @round_v16f32(<16 x float> %x) {
; SSE2-LABEL: round_v16f32:
; SSE2: # %bb.0:
-; SSE2-NEXT: subq $104, %rsp
-; SSE2-NEXT: .cfi_def_cfa_offset 112
-; SSE2-NEXT: movaps %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps %xmm1, (%rsp) # 16-byte Spill
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: movhlps {{.*#+}} xmm0 = xmm0[1,1]
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: unpcklps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; SSE2-NEXT: # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE2-NEXT: unpcklps {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSE2-NEXT: unpcklpd {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE2-NEXT: # xmm1 = xmm1[0],mem[0]
-; SSE2-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps (%rsp), %xmm0 # 16-byte Reload
-; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps (%rsp), %xmm0 # 16-byte Reload
-; SSE2-NEXT: movhlps {{.*#+}} xmm0 = xmm0[1,1]
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: unpcklps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; SSE2-NEXT: # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps (%rsp), %xmm0 # 16-byte Reload
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps (%rsp), %xmm0 # 16-byte Reload
-; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE2-NEXT: unpcklps {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSE2-NEXT: unpcklpd {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE2-NEXT: # xmm1 = xmm1[0],mem[0]
-; SSE2-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: movhlps {{.*#+}} xmm0 = xmm0[1,1]
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: unpcklps (%rsp), %xmm0 # 16-byte Folded Reload
-; SSE2-NEXT: # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: movaps (%rsp), %xmm1 # 16-byte Reload
-; SSE2-NEXT: unpcklps {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSE2-NEXT: unpcklpd {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE2-NEXT: # xmm1 = xmm1[0],mem[0]
-; SSE2-NEXT: movaps %xmm1, (%rsp) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: movhlps {{.*#+}} xmm0 = xmm0[1,1]
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: unpcklps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; SSE2-NEXT: # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
-; SSE2-NEXT: callq roundf at PLT
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; SSE2-NEXT: unpcklps {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1]
-; SSE2-NEXT: unpcklpd {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; SSE2-NEXT: # xmm3 = xmm3[0],mem[0]
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE2-NEXT: movaps (%rsp), %xmm2 # 16-byte Reload
-; SSE2-NEXT: addq $104, %rsp
-; SSE2-NEXT: .cfi_def_cfa_offset 8
+; SSE2-NEXT: movaps {{.*#+}} xmm5 = [NaN,NaN,NaN,NaN]
+; SSE2-NEXT: movaps %xmm0, %xmm4
+; SSE2-NEXT: andps %xmm5, %xmm4
+; SSE2-NEXT: movaps {{.*#+}} xmm7 = [4.9999997E-1,4.9999997E-1,4.9999997E-1,4.9999997E-1]
+; SSE2-NEXT: movaps %xmm4, %xmm6
+; SSE2-NEXT: addps %xmm7, %xmm6
+; SSE2-NEXT: cvttps2dq %xmm6, %xmm6
+; SSE2-NEXT: cvtdq2ps %xmm6, %xmm6
+; SSE2-NEXT: andps %xmm5, %xmm6
+; SSE2-NEXT: movaps %xmm5, %xmm9
+; SSE2-NEXT: movaps %xmm1, %xmm8
+; SSE2-NEXT: andps %xmm5, %xmm8
+; SSE2-NEXT: movaps %xmm8, %xmm10
+; SSE2-NEXT: addps %xmm7, %xmm10
+; SSE2-NEXT: cvttps2dq %xmm10, %xmm10
+; SSE2-NEXT: cvtdq2ps %xmm10, %xmm10
+; SSE2-NEXT: andps %xmm5, %xmm10
+; SSE2-NEXT: movaps %xmm5, %xmm12
+; SSE2-NEXT: movaps %xmm2, %xmm11
+; SSE2-NEXT: andps %xmm5, %xmm11
+; SSE2-NEXT: movaps %xmm11, %xmm13
+; SSE2-NEXT: addps %xmm7, %xmm13
+; SSE2-NEXT: cvttps2dq %xmm13, %xmm13
+; SSE2-NEXT: cvtdq2ps %xmm13, %xmm13
+; SSE2-NEXT: andps %xmm5, %xmm13
+; SSE2-NEXT: movaps %xmm5, %xmm15
+; SSE2-NEXT: movaps %xmm3, %xmm14
+; SSE2-NEXT: andps %xmm5, %xmm14
+; SSE2-NEXT: addps %xmm14, %xmm7
+; SSE2-NEXT: cvttps2dq %xmm7, %xmm7
+; SSE2-NEXT: cvtdq2ps %xmm7, %xmm7
+; SSE2-NEXT: andps %xmm5, %xmm7
+; SSE2-NEXT: andnps %xmm0, %xmm5
+; SSE2-NEXT: orps %xmm5, %xmm6
+; SSE2-NEXT: movdqa {{.*#+}} xmm5 = [1325400063,1325400063,1325400063,1325400063]
+; SSE2-NEXT: pcmpgtd %xmm5, %xmm4
+; SSE2-NEXT: andps %xmm4, %xmm0
+; SSE2-NEXT: andnps %xmm6, %xmm4
+; SSE2-NEXT: orps %xmm4, %xmm0
+; SSE2-NEXT: andnps %xmm1, %xmm9
+; SSE2-NEXT: orps %xmm9, %xmm10
+; SSE2-NEXT: pcmpgtd %xmm5, %xmm8
+; SSE2-NEXT: andps %xmm8, %xmm1
+; SSE2-NEXT: andnps %xmm10, %xmm8
+; SSE2-NEXT: orps %xmm8, %xmm1
+; SSE2-NEXT: andnps %xmm2, %xmm12
+; SSE2-NEXT: orps %xmm12, %xmm13
+; SSE2-NEXT: pcmpgtd %xmm5, %xmm11
+; SSE2-NEXT: andps %xmm11, %xmm2
+; SSE2-NEXT: andnps %xmm13, %xmm11
+; SSE2-NEXT: orps %xmm11, %xmm2
+; SSE2-NEXT: andnps %xmm3, %xmm15
+; SSE2-NEXT: orps %xmm15, %xmm7
+; SSE2-NEXT: pcmpgtd %xmm5, %xmm14
+; SSE2-NEXT: andps %xmm14, %xmm3
+; SSE2-NEXT: andnps %xmm7, %xmm14
+; SSE2-NEXT: orps %xmm14, %xmm3
; SSE2-NEXT: retq
;
; SSE41-LABEL: round_v16f32:
@@ -601,51 +606,87 @@ define <16 x float> @round_v16f32(<16 x float> %x) {
define <8 x double> @round_v8f64(<8 x double> %x) {
; SSE2-LABEL: round_v8f64:
; SSE2: # %bb.0:
-; SSE2-NEXT: subq $88, %rsp
-; SSE2-NEXT: .cfi_def_cfa_offset 96
-; SSE2-NEXT: movaps %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps %xmm1, (%rsp) # 16-byte Spill
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: callq round at PLT
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: movhlps {{.*#+}} xmm0 = xmm0[1,1]
-; SSE2-NEXT: callq round at PLT
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE2-NEXT: movlhps {{.*#+}} xmm1 = xmm1[0],xmm0[0]
-; SSE2-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps (%rsp), %xmm0 # 16-byte Reload
-; SSE2-NEXT: callq round at PLT
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps (%rsp), %xmm0 # 16-byte Reload
-; SSE2-NEXT: movhlps {{.*#+}} xmm0 = xmm0[1,1]
-; SSE2-NEXT: callq round at PLT
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE2-NEXT: movlhps {{.*#+}} xmm1 = xmm1[0],xmm0[0]
-; SSE2-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: callq round at PLT
-; SSE2-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: movhlps {{.*#+}} xmm0 = xmm0[1,1]
-; SSE2-NEXT: callq round at PLT
-; SSE2-NEXT: movaps (%rsp), %xmm1 # 16-byte Reload
-; SSE2-NEXT: movlhps {{.*#+}} xmm1 = xmm1[0],xmm0[0]
-; SSE2-NEXT: movaps %xmm1, (%rsp) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: callq round at PLT
-; SSE2-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: movhlps {{.*#+}} xmm0 = xmm0[1,1]
-; SSE2-NEXT: callq round at PLT
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; SSE2-NEXT: movlhps {{.*#+}} xmm3 = xmm3[0],xmm0[0]
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE2-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE2-NEXT: movaps (%rsp), %xmm2 # 16-byte Reload
-; SSE2-NEXT: addq $88, %rsp
-; SSE2-NEXT: .cfi_def_cfa_offset 8
+; SSE2-NEXT: movdqa {{.*#+}} xmm4 = [NaN,NaN]
+; SSE2-NEXT: movdqa %xmm0, %xmm6
+; SSE2-NEXT: pand %xmm4, %xmm6
+; SSE2-NEXT: movapd {{.*#+}} xmm5 = [4.9999999999999994E-1,4.9999999999999994E-1]
+; SSE2-NEXT: pshufd {{.*#+}} xmm7 = xmm6[1,1,3,3]
+; SSE2-NEXT: addpd %xmm5, %xmm6
+; SSE2-NEXT: cvttsd2si %xmm6, %rax
+; SSE2-NEXT: cvtsi2sd %rax, %xmm8
+; SSE2-NEXT: unpckhpd {{.*#+}} xmm6 = xmm6[1,1]
+; SSE2-NEXT: cvttsd2si %xmm6, %rax
+; SSE2-NEXT: xorps %xmm6, %xmm6
+; SSE2-NEXT: cvtsi2sd %rax, %xmm6
+; SSE2-NEXT: unpcklpd {{.*#+}} xmm8 = xmm8[0],xmm6[0]
+; SSE2-NEXT: andpd %xmm4, %xmm8
+; SSE2-NEXT: movdqa %xmm4, %xmm9
+; SSE2-NEXT: pandn %xmm0, %xmm9
+; SSE2-NEXT: por %xmm8, %xmm9
+; SSE2-NEXT: movdqa {{.*#+}} xmm6 = [1127219199,1127219199,1127219199,1127219199]
+; SSE2-NEXT: pcmpgtd %xmm6, %xmm7
+; SSE2-NEXT: pand %xmm7, %xmm0
+; SSE2-NEXT: pandn %xmm9, %xmm7
+; SSE2-NEXT: por %xmm7, %xmm0
+; SSE2-NEXT: movdqa %xmm1, %xmm7
+; SSE2-NEXT: pand %xmm4, %xmm7
+; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm7[1,1,3,3]
+; SSE2-NEXT: addpd %xmm5, %xmm7
+; SSE2-NEXT: cvttsd2si %xmm7, %rax
+; SSE2-NEXT: xorps %xmm9, %xmm9
+; SSE2-NEXT: cvtsi2sd %rax, %xmm9
+; SSE2-NEXT: unpckhpd {{.*#+}} xmm7 = xmm7[1,1]
+; SSE2-NEXT: cvttsd2si %xmm7, %rax
+; SSE2-NEXT: xorps %xmm7, %xmm7
+; SSE2-NEXT: cvtsi2sd %rax, %xmm7
+; SSE2-NEXT: unpcklpd {{.*#+}} xmm9 = xmm9[0],xmm7[0]
+; SSE2-NEXT: andpd %xmm4, %xmm9
+; SSE2-NEXT: movdqa %xmm4, %xmm7
+; SSE2-NEXT: pandn %xmm1, %xmm7
+; SSE2-NEXT: por %xmm9, %xmm7
+; SSE2-NEXT: pcmpgtd %xmm6, %xmm8
+; SSE2-NEXT: pand %xmm8, %xmm1
+; SSE2-NEXT: pandn %xmm7, %xmm8
+; SSE2-NEXT: por %xmm8, %xmm1
+; SSE2-NEXT: movdqa %xmm2, %xmm7
+; SSE2-NEXT: pand %xmm4, %xmm7
+; SSE2-NEXT: pshufd {{.*#+}} xmm8 = xmm7[1,1,3,3]
+; SSE2-NEXT: addpd %xmm5, %xmm7
+; SSE2-NEXT: cvttsd2si %xmm7, %rax
+; SSE2-NEXT: xorps %xmm9, %xmm9
+; SSE2-NEXT: cvtsi2sd %rax, %xmm9
+; SSE2-NEXT: unpckhpd {{.*#+}} xmm7 = xmm7[1,1]
+; SSE2-NEXT: cvttsd2si %xmm7, %rax
+; SSE2-NEXT: xorps %xmm7, %xmm7
+; SSE2-NEXT: cvtsi2sd %rax, %xmm7
+; SSE2-NEXT: unpcklpd {{.*#+}} xmm9 = xmm9[0],xmm7[0]
+; SSE2-NEXT: andpd %xmm4, %xmm9
+; SSE2-NEXT: movdqa %xmm4, %xmm7
+; SSE2-NEXT: pandn %xmm2, %xmm7
+; SSE2-NEXT: por %xmm9, %xmm7
+; SSE2-NEXT: pcmpgtd %xmm6, %xmm8
+; SSE2-NEXT: pand %xmm8, %xmm2
+; SSE2-NEXT: pandn %xmm7, %xmm8
+; SSE2-NEXT: por %xmm8, %xmm2
+; SSE2-NEXT: movapd %xmm3, %xmm7
+; SSE2-NEXT: andpd %xmm4, %xmm7
+; SSE2-NEXT: addpd %xmm7, %xmm5
+; SSE2-NEXT: cvttsd2si %xmm5, %rax
+; SSE2-NEXT: xorps %xmm8, %xmm8
+; SSE2-NEXT: cvtsi2sd %rax, %xmm8
+; SSE2-NEXT: unpckhpd {{.*#+}} xmm5 = xmm5[1,1]
+; SSE2-NEXT: cvttsd2si %xmm5, %rax
+; SSE2-NEXT: xorps %xmm5, %xmm5
+; SSE2-NEXT: cvtsi2sd %rax, %xmm5
+; SSE2-NEXT: unpcklpd {{.*#+}} xmm8 = xmm8[0],xmm5[0]
+; SSE2-NEXT: andpd %xmm4, %xmm8
+; SSE2-NEXT: andnpd %xmm3, %xmm4
+; SSE2-NEXT: orpd %xmm8, %xmm4
+; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm7[1,1,3,3]
+; SSE2-NEXT: pcmpgtd %xmm6, %xmm5
+; SSE2-NEXT: andpd %xmm5, %xmm3
+; SSE2-NEXT: andnpd %xmm4, %xmm5
+; SSE2-NEXT: orpd %xmm5, %xmm3
; SSE2-NEXT: retq
;
; SSE41-LABEL: round_v8f64:
diff --git a/llvm/test/CodeGen/X86/fp16-libcalls.ll b/llvm/test/CodeGen/X86/fp16-libcalls.ll
index 6abdf9a5ba652..594418e9d4b0b 100644
--- a/llvm/test/CodeGen/X86/fp16-libcalls.ll
+++ b/llvm/test/CodeGen/X86/fp16-libcalls.ll
@@ -1152,7 +1152,20 @@ define void @test_half_trunc(half %a0, ptr %p0) nounwind {
; X64-NEXT: pushq %rbx
; X64-NEXT: movq %rdi, %rbx
; X64-NEXT: callq __extendhfsf2 at PLT
-; X64-NEXT: callq truncf at PLT
+; X64-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN]
+; X64-NEXT: movdqa %xmm0, %xmm2
+; X64-NEXT: pand %xmm1, %xmm2
+; X64-NEXT: movd %xmm2, %eax
+; X64-NEXT: cmpl $1325400064, %eax # imm = 0x4F000000
+; X64-NEXT: jge .LBB20_2
+; X64-NEXT: # %bb.1:
+; X64-NEXT: cvttps2dq %xmm2, %xmm2
+; X64-NEXT: cvtdq2ps %xmm2, %xmm2
+; X64-NEXT: andps %xmm1, %xmm2
+; X64-NEXT: pandn %xmm0, %xmm1
+; X64-NEXT: por %xmm2, %xmm1
+; X64-NEXT: movdqa %xmm1, %xmm0
+; X64-NEXT: .LBB20_2:
; X64-NEXT: callq __truncsfhf2 at PLT
; X64-NEXT: pextrw $0, %xmm0, %eax
; X64-NEXT: movw %ax, (%rbx)
@@ -1168,9 +1181,23 @@ define void @test_half_trunc(half %a0, ptr %p0) nounwind {
; X86-NEXT: pextrw $0, %xmm0, %eax
; X86-NEXT: movw %ax, (%esp)
; X86-NEXT: calll __extendhfsf2
-; X86-NEXT: fstps (%esp)
-; X86-NEXT: calll truncf
-; X86-NEXT: fstps (%esp)
+; X86-NEXT: fstps {{[0-9]+}}(%esp)
+; X86-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X86-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN]
+; X86-NEXT: movdqa %xmm0, %xmm2
+; X86-NEXT: pand %xmm1, %xmm2
+; X86-NEXT: movd %xmm2, %eax
+; X86-NEXT: cmpl $1325400064, %eax # imm = 0x4F000000
+; X86-NEXT: jge .LBB20_2
+; X86-NEXT: # %bb.1:
+; X86-NEXT: cvttps2dq %xmm2, %xmm2
+; X86-NEXT: cvtdq2ps %xmm2, %xmm2
+; X86-NEXT: andps %xmm1, %xmm2
+; X86-NEXT: pandn %xmm0, %xmm1
+; X86-NEXT: por %xmm2, %xmm1
+; X86-NEXT: movdqa %xmm1, %xmm0
+; X86-NEXT: .LBB20_2:
+; X86-NEXT: movd %xmm0, (%esp)
; X86-NEXT: calll __truncsfhf2
; X86-NEXT: pextrw $0, %xmm0, %eax
; X86-NEXT: movw %ax, (%esi)
diff --git a/llvm/test/CodeGen/X86/freeze-unary.ll b/llvm/test/CodeGen/X86/freeze-unary.ll
index 222ec1eb3b9de..addae1acca77a 100644
--- a/llvm/test/CodeGen/X86/freeze-unary.ll
+++ b/llvm/test/CodeGen/X86/freeze-unary.ll
@@ -485,9 +485,24 @@ define float @ftrunc_freeze_fround(float %a0) nounwind {
; X86-LABEL: ftrunc_freeze_fround:
; X86: # %bb.0:
; X86-NEXT: pushl %eax
-; X86-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-NEXT: movss %xmm0, (%esp)
-; X86-NEXT: calll roundf
+; X86-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X86-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN]
+; X86-NEXT: movdqa %xmm0, %xmm2
+; X86-NEXT: pand %xmm1, %xmm2
+; X86-NEXT: movd %xmm2, %eax
+; X86-NEXT: cmpl $1325400064, %eax # imm = 0x4F000000
+; X86-NEXT: jge .LBB23_2
+; X86-NEXT: # %bb.1:
+; X86-NEXT: addss {{\.?LCPI[0-9]+_[0-9]+}}, %xmm2
+; X86-NEXT: cvttps2dq %xmm2, %xmm2
+; X86-NEXT: cvtdq2ps %xmm2, %xmm2
+; X86-NEXT: andps %xmm1, %xmm2
+; X86-NEXT: pandn %xmm0, %xmm1
+; X86-NEXT: por %xmm2, %xmm1
+; X86-NEXT: movdqa %xmm1, %xmm0
+; X86-NEXT: .LBB23_2:
+; X86-NEXT: movd %xmm0, (%esp)
+; X86-NEXT: flds (%esp)
; X86-NEXT: popl %eax
; X86-NEXT: retl
;
@@ -550,9 +565,23 @@ define float @ftrunc_freeze_ftrunc(float %a0) nounwind {
; X86-LABEL: ftrunc_freeze_ftrunc:
; X86: # %bb.0:
; X86-NEXT: pushl %eax
-; X86-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-NEXT: movss %xmm0, (%esp)
-; X86-NEXT: calll truncf
+; X86-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X86-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN]
+; X86-NEXT: movdqa %xmm0, %xmm2
+; X86-NEXT: pand %xmm1, %xmm2
+; X86-NEXT: movd %xmm2, %eax
+; X86-NEXT: cmpl $1325400064, %eax # imm = 0x4F000000
+; X86-NEXT: jge .LBB26_2
+; X86-NEXT: # %bb.1:
+; X86-NEXT: cvttps2dq %xmm2, %xmm2
+; X86-NEXT: cvtdq2ps %xmm2, %xmm2
+; X86-NEXT: andps %xmm1, %xmm2
+; X86-NEXT: pandn %xmm0, %xmm1
+; X86-NEXT: por %xmm2, %xmm1
+; X86-NEXT: movdqa %xmm1, %xmm0
+; X86-NEXT: .LBB26_2:
+; X86-NEXT: movd %xmm0, (%esp)
+; X86-NEXT: flds (%esp)
; X86-NEXT: popl %eax
; X86-NEXT: retl
;
diff --git a/llvm/test/CodeGen/X86/ftrunc.ll b/llvm/test/CodeGen/X86/ftrunc.ll
index 1ab9040a2810f..3e44569a33757 100644
--- a/llvm/test/CodeGen/X86/ftrunc.ll
+++ b/llvm/test/CodeGen/X86/ftrunc.ll
@@ -10,10 +10,20 @@ declare i64 @llvm.fptosi.sat.i64.f64(double)
define float @trunc_unsigned_f32(float %x) #0 {
; SSE2-LABEL: trunc_unsigned_f32:
; SSE2: # %bb.0:
-; SSE2-NEXT: cvttss2si %xmm0, %rax
-; SSE2-NEXT: movl %eax, %eax
-; SSE2-NEXT: xorps %xmm0, %xmm0
-; SSE2-NEXT: cvtsi2ss %rax, %xmm0
+; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN]
+; SSE2-NEXT: movdqa %xmm0, %xmm2
+; SSE2-NEXT: pand %xmm1, %xmm2
+; SSE2-NEXT: movd %xmm2, %eax
+; SSE2-NEXT: cmpl $1325400064, %eax # imm = 0x4F000000
+; SSE2-NEXT: jge .LBB0_2
+; SSE2-NEXT: # %bb.1:
+; SSE2-NEXT: cvttps2dq %xmm2, %xmm2
+; SSE2-NEXT: cvtdq2ps %xmm2, %xmm2
+; SSE2-NEXT: andps %xmm1, %xmm2
+; SSE2-NEXT: pandn %xmm0, %xmm1
+; SSE2-NEXT: por %xmm2, %xmm1
+; SSE2-NEXT: movdqa %xmm1, %xmm0
+; SSE2-NEXT: .LBB0_2:
; SSE2-NEXT: retq
;
; SSE41-LABEL: trunc_unsigned_f32:
@@ -43,19 +53,22 @@ define float @trunc_unsigned_f32(float %x) #0 {
define double @trunc_unsigned_f64(double %x) #0 {
; SSE2-LABEL: trunc_unsigned_f64:
; SSE2: # %bb.0:
-; SSE2-NEXT: cvttsd2si %xmm0, %rax
-; SSE2-NEXT: movq %rax, %rcx
-; SSE2-NEXT: sarq $63, %rcx
-; SSE2-NEXT: subsd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; SSE2-NEXT: cvttsd2si %xmm0, %rdx
-; SSE2-NEXT: andq %rcx, %rdx
-; SSE2-NEXT: orq %rax, %rdx
-; SSE2-NEXT: movq %rdx, %xmm1
-; SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[1],mem[1]
-; SSE2-NEXT: subpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; SSE2-NEXT: movapd %xmm1, %xmm0
-; SSE2-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
-; SSE2-NEXT: addsd %xmm1, %xmm0
+; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN]
+; SSE2-NEXT: movdqa %xmm0, %xmm2
+; SSE2-NEXT: pand %xmm1, %xmm2
+; SSE2-NEXT: movq %xmm2, %rax
+; SSE2-NEXT: movabsq $4841369599423283199, %rcx # imm = 0x432FFFFFFFFFFFFF
+; SSE2-NEXT: cmpq %rcx, %rax
+; SSE2-NEXT: jg .LBB1_2
+; SSE2-NEXT: # %bb.1:
+; SSE2-NEXT: cvttsd2si %xmm2, %rax
+; SSE2-NEXT: xorps %xmm2, %xmm2
+; SSE2-NEXT: cvtsi2sd %rax, %xmm2
+; SSE2-NEXT: andpd %xmm1, %xmm2
+; SSE2-NEXT: pandn %xmm0, %xmm1
+; SSE2-NEXT: por %xmm2, %xmm1
+; SSE2-NEXT: movdqa %xmm1, %xmm0
+; SSE2-NEXT: .LBB1_2:
; SSE2-NEXT: retq
;
; SSE41-LABEL: trunc_unsigned_f64:
@@ -89,20 +102,18 @@ define double @trunc_unsigned_f64(double %x) #0 {
define <4 x float> @trunc_unsigned_v4f32(<4 x float> %x) #0 {
; SSE2-LABEL: trunc_unsigned_v4f32:
; SSE2: # %bb.0:
-; SSE2-NEXT: cvttps2dq %xmm0, %xmm1
-; SSE2-NEXT: movdqa %xmm1, %xmm2
-; SSE2-NEXT: psrad $31, %xmm2
-; SSE2-NEXT: subps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; SSE2-NEXT: cvttps2dq %xmm0, %xmm0
-; SSE2-NEXT: pand %xmm2, %xmm0
-; SSE2-NEXT: por %xmm1, %xmm0
-; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [65535,65535,65535,65535]
-; SSE2-NEXT: pand %xmm0, %xmm1
-; SSE2-NEXT: por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; SSE2-NEXT: psrld $16, %xmm0
-; SSE2-NEXT: por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; SSE2-NEXT: subps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; SSE2-NEXT: addps %xmm1, %xmm0
+; SSE2-NEXT: movaps {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN]
+; SSE2-NEXT: movaps %xmm0, %xmm2
+; SSE2-NEXT: andps %xmm1, %xmm2
+; SSE2-NEXT: cvttps2dq %xmm2, %xmm3
+; SSE2-NEXT: cvtdq2ps %xmm3, %xmm3
+; SSE2-NEXT: andps %xmm1, %xmm3
+; SSE2-NEXT: andnps %xmm0, %xmm1
+; SSE2-NEXT: orps %xmm1, %xmm3
+; SSE2-NEXT: pcmpgtd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; SSE2-NEXT: andps %xmm2, %xmm0
+; SSE2-NEXT: andnps %xmm3, %xmm2
+; SSE2-NEXT: orps %xmm2, %xmm0
; SSE2-NEXT: retq
;
; SSE41-LABEL: trunc_unsigned_v4f32:
@@ -122,34 +133,24 @@ define <4 x float> @trunc_unsigned_v4f32(<4 x float> %x) #0 {
define <2 x double> @trunc_unsigned_v2f64(<2 x double> %x) #0 {
; SSE2-LABEL: trunc_unsigned_v2f64:
; SSE2: # %bb.0:
-; SSE2-NEXT: movsd {{.*#+}} xmm2 = [9.2233720368547758E+18,0.0E+0]
-; SSE2-NEXT: movapd %xmm0, %xmm1
-; SSE2-NEXT: subsd %xmm2, %xmm1
-; SSE2-NEXT: cvttsd2si %xmm1, %rax
-; SSE2-NEXT: cvttsd2si %xmm0, %rcx
-; SSE2-NEXT: movq %rcx, %rdx
-; SSE2-NEXT: sarq $63, %rdx
-; SSE2-NEXT: andq %rax, %rdx
-; SSE2-NEXT: orq %rcx, %rdx
-; SSE2-NEXT: movq %rdx, %xmm1
-; SSE2-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1,1]
-; SSE2-NEXT: cvttsd2si %xmm0, %rax
-; SSE2-NEXT: subsd %xmm2, %xmm0
-; SSE2-NEXT: cvttsd2si %xmm0, %rcx
-; SSE2-NEXT: movq %rax, %rdx
-; SSE2-NEXT: sarq $63, %rdx
-; SSE2-NEXT: andq %rcx, %rdx
-; SSE2-NEXT: orq %rax, %rdx
-; SSE2-NEXT: movq %rdx, %xmm0
-; SSE2-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
-; SSE2-NEXT: movdqa {{.*#+}} xmm0 = [4294967295,4294967295]
-; SSE2-NEXT: pand %xmm1, %xmm0
-; SSE2-NEXT: por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; SSE2-NEXT: psrlq $32, %xmm1
-; SSE2-NEXT: por {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; SSE2-NEXT: subpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; SSE2-NEXT: addpd %xmm0, %xmm1
-; SSE2-NEXT: movapd %xmm1, %xmm0
+; SSE2-NEXT: movapd {{.*#+}} xmm1 = [NaN,NaN]
+; SSE2-NEXT: movapd %xmm0, %xmm2
+; SSE2-NEXT: andpd %xmm1, %xmm2
+; SSE2-NEXT: cvttsd2si %xmm2, %rax
+; SSE2-NEXT: cvtsi2sd %rax, %xmm3
+; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm2[2,3,2,3]
+; SSE2-NEXT: cvttsd2si %xmm4, %rax
+; SSE2-NEXT: xorps %xmm4, %xmm4
+; SSE2-NEXT: cvtsi2sd %rax, %xmm4
+; SSE2-NEXT: unpcklpd {{.*#+}} xmm3 = xmm3[0],xmm4[0]
+; SSE2-NEXT: andpd %xmm1, %xmm3
+; SSE2-NEXT: andnpd %xmm0, %xmm1
+; SSE2-NEXT: orpd %xmm3, %xmm1
+; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; SSE2-NEXT: pcmpgtd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; SSE2-NEXT: andpd %xmm2, %xmm0
+; SSE2-NEXT: andnpd %xmm1, %xmm2
+; SSE2-NEXT: orpd %xmm2, %xmm0
; SSE2-NEXT: retq
;
; SSE41-LABEL: trunc_unsigned_v2f64:
@@ -169,63 +170,44 @@ define <2 x double> @trunc_unsigned_v2f64(<2 x double> %x) #0 {
define <4 x double> @trunc_unsigned_v4f64(<4 x double> %x) #0 {
; SSE2-LABEL: trunc_unsigned_v4f64:
; SSE2: # %bb.0:
-; SSE2-NEXT: movapd %xmm1, %xmm2
-; SSE2-NEXT: movsd {{.*#+}} xmm3 = [9.2233720368547758E+18,0.0E+0]
-; SSE2-NEXT: subsd %xmm3, %xmm1
-; SSE2-NEXT: cvttsd2si %xmm1, %rax
-; SSE2-NEXT: cvttsd2si %xmm2, %rcx
-; SSE2-NEXT: movq %rcx, %rdx
-; SSE2-NEXT: sarq $63, %rdx
-; SSE2-NEXT: andq %rax, %rdx
-; SSE2-NEXT: orq %rcx, %rdx
-; SSE2-NEXT: movq %rdx, %xmm1
-; SSE2-NEXT: unpckhpd {{.*#+}} xmm2 = xmm2[1,1]
-; SSE2-NEXT: cvttsd2si %xmm2, %rax
-; SSE2-NEXT: subsd %xmm3, %xmm2
-; SSE2-NEXT: cvttsd2si %xmm2, %rcx
-; SSE2-NEXT: movq %rax, %rdx
-; SSE2-NEXT: sarq $63, %rdx
-; SSE2-NEXT: andq %rcx, %rdx
-; SSE2-NEXT: orq %rax, %rdx
-; SSE2-NEXT: movq %rdx, %xmm2
-; SSE2-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
-; SSE2-NEXT: movapd %xmm0, %xmm2
-; SSE2-NEXT: subsd %xmm3, %xmm2
-; SSE2-NEXT: cvttsd2si %xmm2, %rax
-; SSE2-NEXT: cvttsd2si %xmm0, %rcx
-; SSE2-NEXT: movq %rcx, %rdx
-; SSE2-NEXT: sarq $63, %rdx
-; SSE2-NEXT: andq %rax, %rdx
-; SSE2-NEXT: orq %rcx, %rdx
-; SSE2-NEXT: movq %rdx, %xmm2
-; SSE2-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1,1]
-; SSE2-NEXT: cvttsd2si %xmm0, %rax
-; SSE2-NEXT: subsd %xmm3, %xmm0
-; SSE2-NEXT: cvttsd2si %xmm0, %rcx
-; SSE2-NEXT: movq %rax, %rdx
-; SSE2-NEXT: sarq $63, %rdx
-; SSE2-NEXT: andq %rcx, %rdx
-; SSE2-NEXT: orq %rax, %rdx
-; SSE2-NEXT: movq %rdx, %xmm0
-; SSE2-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm0[0]
-; SSE2-NEXT: movdqa {{.*#+}} xmm0 = [4294967295,4294967295]
-; SSE2-NEXT: movdqa %xmm2, %xmm3
-; SSE2-NEXT: pand %xmm0, %xmm3
-; SSE2-NEXT: movdqa {{.*#+}} xmm4 = [4841369599423283200,4841369599423283200]
-; SSE2-NEXT: por %xmm4, %xmm3
-; SSE2-NEXT: psrlq $32, %xmm2
-; SSE2-NEXT: movdqa {{.*#+}} xmm5 = [4985484787499139072,4985484787499139072]
-; SSE2-NEXT: por %xmm5, %xmm2
-; SSE2-NEXT: movapd {{.*#+}} xmm6 = [1.9342813118337666E+25,1.9342813118337666E+25]
-; SSE2-NEXT: subpd %xmm6, %xmm2
-; SSE2-NEXT: addpd %xmm3, %xmm2
-; SSE2-NEXT: pand %xmm1, %xmm0
-; SSE2-NEXT: por %xmm4, %xmm0
-; SSE2-NEXT: psrlq $32, %xmm1
-; SSE2-NEXT: por %xmm5, %xmm1
-; SSE2-NEXT: subpd %xmm6, %xmm1
-; SSE2-NEXT: addpd %xmm0, %xmm1
-; SSE2-NEXT: movapd %xmm2, %xmm0
+; SSE2-NEXT: movapd {{.*#+}} xmm2 = [NaN,NaN]
+; SSE2-NEXT: movapd %xmm0, %xmm3
+; SSE2-NEXT: andpd %xmm2, %xmm3
+; SSE2-NEXT: cvttsd2si %xmm3, %rax
+; SSE2-NEXT: cvtsi2sd %rax, %xmm4
+; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm3[2,3,2,3]
+; SSE2-NEXT: cvttsd2si %xmm5, %rax
+; SSE2-NEXT: xorps %xmm5, %xmm5
+; SSE2-NEXT: cvtsi2sd %rax, %xmm5
+; SSE2-NEXT: unpcklpd {{.*#+}} xmm4 = xmm4[0],xmm5[0]
+; SSE2-NEXT: andpd %xmm2, %xmm4
+; SSE2-NEXT: movapd %xmm2, %xmm5
+; SSE2-NEXT: andnpd %xmm0, %xmm5
+; SSE2-NEXT: orpd %xmm4, %xmm5
+; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
+; SSE2-NEXT: movdqa {{.*#+}} xmm4 = [1127219199,1127219199,1127219199,1127219199]
+; SSE2-NEXT: pcmpgtd %xmm4, %xmm3
+; SSE2-NEXT: andpd %xmm3, %xmm0
+; SSE2-NEXT: andnpd %xmm5, %xmm3
+; SSE2-NEXT: orpd %xmm3, %xmm0
+; SSE2-NEXT: movapd %xmm1, %xmm3
+; SSE2-NEXT: andpd %xmm2, %xmm3
+; SSE2-NEXT: cvttsd2si %xmm3, %rax
+; SSE2-NEXT: xorps %xmm5, %xmm5
+; SSE2-NEXT: cvtsi2sd %rax, %xmm5
+; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm3[2,3,2,3]
+; SSE2-NEXT: cvttsd2si %xmm6, %rax
+; SSE2-NEXT: xorps %xmm6, %xmm6
+; SSE2-NEXT: cvtsi2sd %rax, %xmm6
+; SSE2-NEXT: unpcklpd {{.*#+}} xmm5 = xmm5[0],xmm6[0]
+; SSE2-NEXT: andpd %xmm2, %xmm5
+; SSE2-NEXT: andnpd %xmm1, %xmm2
+; SSE2-NEXT: orpd %xmm5, %xmm2
+; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
+; SSE2-NEXT: pcmpgtd %xmm4, %xmm3
+; SSE2-NEXT: andpd %xmm3, %xmm1
+; SSE2-NEXT: andnpd %xmm2, %xmm3
+; SSE2-NEXT: orpd %xmm3, %xmm1
; SSE2-NEXT: retq
;
; SSE41-LABEL: trunc_unsigned_v4f64:
@@ -276,8 +258,20 @@ define float @trunc_signed_f32_no_fast_math(float %x) nounwind {
define float @trunc_signed_f32_nsz(float %x) #0 {
; SSE2-LABEL: trunc_signed_f32_nsz:
; SSE2: # %bb.0:
-; SSE2-NEXT: cvttps2dq %xmm0, %xmm0
-; SSE2-NEXT: cvtdq2ps %xmm0, %xmm0
+; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN]
+; SSE2-NEXT: movdqa %xmm0, %xmm2
+; SSE2-NEXT: pand %xmm1, %xmm2
+; SSE2-NEXT: movd %xmm2, %eax
+; SSE2-NEXT: cmpl $1325400064, %eax # imm = 0x4F000000
+; SSE2-NEXT: jge .LBB6_2
+; SSE2-NEXT: # %bb.1:
+; SSE2-NEXT: cvttps2dq %xmm2, %xmm2
+; SSE2-NEXT: cvtdq2ps %xmm2, %xmm2
+; SSE2-NEXT: andps %xmm1, %xmm2
+; SSE2-NEXT: pandn %xmm0, %xmm1
+; SSE2-NEXT: por %xmm2, %xmm1
+; SSE2-NEXT: movdqa %xmm1, %xmm0
+; SSE2-NEXT: .LBB6_2:
; SSE2-NEXT: retq
;
; SSE41-LABEL: trunc_signed_f32_nsz:
@@ -339,8 +333,22 @@ define double @trunc_signed32_f64_no_fast_math(double %x) nounwind {
define double @trunc_signed32_f64_nsz(double %x) #0 {
; SSE2-LABEL: trunc_signed32_f64_nsz:
; SSE2: # %bb.0:
-; SSE2-NEXT: cvttpd2dq %xmm0, %xmm0
-; SSE2-NEXT: cvtdq2pd %xmm0, %xmm0
+; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN]
+; SSE2-NEXT: movdqa %xmm0, %xmm2
+; SSE2-NEXT: pand %xmm1, %xmm2
+; SSE2-NEXT: movq %xmm2, %rax
+; SSE2-NEXT: movabsq $4841369599423283199, %rcx # imm = 0x432FFFFFFFFFFFFF
+; SSE2-NEXT: cmpq %rcx, %rax
+; SSE2-NEXT: jg .LBB8_2
+; SSE2-NEXT: # %bb.1:
+; SSE2-NEXT: cvttsd2si %xmm2, %rax
+; SSE2-NEXT: xorps %xmm2, %xmm2
+; SSE2-NEXT: cvtsi2sd %rax, %xmm2
+; SSE2-NEXT: andpd %xmm1, %xmm2
+; SSE2-NEXT: pandn %xmm0, %xmm1
+; SSE2-NEXT: por %xmm2, %xmm1
+; SSE2-NEXT: movdqa %xmm1, %xmm0
+; SSE2-NEXT: .LBB8_2:
; SSE2-NEXT: retq
;
; SSE41-LABEL: trunc_signed32_f64_nsz:
@@ -531,9 +539,22 @@ define double @trunc_signed_f64_no_fast_math(double %x) nounwind {
define double @trunc_signed_f64_nsz(double %x) #0 {
; SSE2-LABEL: trunc_signed_f64_nsz:
; SSE2: # %bb.0:
-; SSE2-NEXT: cvttsd2si %xmm0, %rax
-; SSE2-NEXT: xorps %xmm0, %xmm0
-; SSE2-NEXT: cvtsi2sd %rax, %xmm0
+; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN]
+; SSE2-NEXT: movdqa %xmm0, %xmm2
+; SSE2-NEXT: pand %xmm1, %xmm2
+; SSE2-NEXT: movq %xmm2, %rax
+; SSE2-NEXT: movabsq $4841369599423283199, %rcx # imm = 0x432FFFFFFFFFFFFF
+; SSE2-NEXT: cmpq %rcx, %rax
+; SSE2-NEXT: jg .LBB14_2
+; SSE2-NEXT: # %bb.1:
+; SSE2-NEXT: cvttsd2si %xmm2, %rax
+; SSE2-NEXT: xorps %xmm2, %xmm2
+; SSE2-NEXT: cvtsi2sd %rax, %xmm2
+; SSE2-NEXT: andpd %xmm1, %xmm2
+; SSE2-NEXT: pandn %xmm0, %xmm1
+; SSE2-NEXT: por %xmm2, %xmm1
+; SSE2-NEXT: movdqa %xmm1, %xmm0
+; SSE2-NEXT: .LBB14_2:
; SSE2-NEXT: retq
;
; SSE41-LABEL: trunc_signed_f64_nsz:
@@ -567,8 +588,18 @@ define double @trunc_signed_f64_nsz(double %x) #0 {
define <4 x float> @trunc_signed_v4f32_nsz(<4 x float> %x) #0 {
; SSE2-LABEL: trunc_signed_v4f32_nsz:
; SSE2: # %bb.0:
-; SSE2-NEXT: cvttps2dq %xmm0, %xmm0
-; SSE2-NEXT: cvtdq2ps %xmm0, %xmm0
+; SSE2-NEXT: movaps {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN]
+; SSE2-NEXT: movaps %xmm0, %xmm2
+; SSE2-NEXT: andps %xmm1, %xmm2
+; SSE2-NEXT: cvttps2dq %xmm2, %xmm3
+; SSE2-NEXT: cvtdq2ps %xmm3, %xmm3
+; SSE2-NEXT: andps %xmm1, %xmm3
+; SSE2-NEXT: andnps %xmm0, %xmm1
+; SSE2-NEXT: orps %xmm1, %xmm3
+; SSE2-NEXT: pcmpgtd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; SSE2-NEXT: andps %xmm2, %xmm0
+; SSE2-NEXT: andnps %xmm3, %xmm2
+; SSE2-NEXT: orps %xmm2, %xmm0
; SSE2-NEXT: retq
;
; SSE41-LABEL: trunc_signed_v4f32_nsz:
@@ -588,13 +619,24 @@ define <4 x float> @trunc_signed_v4f32_nsz(<4 x float> %x) #0 {
define <2 x double> @trunc_signed_v2f64_nsz(<2 x double> %x) #0 {
; SSE2-LABEL: trunc_signed_v2f64_nsz:
; SSE2: # %bb.0:
-; SSE2-NEXT: cvttsd2si %xmm0, %rax
-; SSE2-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1,1]
-; SSE2-NEXT: cvttsd2si %xmm0, %rcx
-; SSE2-NEXT: xorps %xmm0, %xmm0
-; SSE2-NEXT: cvtsi2sd %rax, %xmm0
-; SSE2-NEXT: cvtsi2sd %rcx, %xmm1
-; SSE2-NEXT: unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSE2-NEXT: movapd {{.*#+}} xmm1 = [NaN,NaN]
+; SSE2-NEXT: movapd %xmm0, %xmm2
+; SSE2-NEXT: andpd %xmm1, %xmm2
+; SSE2-NEXT: cvttsd2si %xmm2, %rax
+; SSE2-NEXT: cvtsi2sd %rax, %xmm3
+; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm2[2,3,2,3]
+; SSE2-NEXT: cvttsd2si %xmm4, %rax
+; SSE2-NEXT: xorps %xmm4, %xmm4
+; SSE2-NEXT: cvtsi2sd %rax, %xmm4
+; SSE2-NEXT: unpcklpd {{.*#+}} xmm3 = xmm3[0],xmm4[0]
+; SSE2-NEXT: andpd %xmm1, %xmm3
+; SSE2-NEXT: andnpd %xmm0, %xmm1
+; SSE2-NEXT: orpd %xmm3, %xmm1
+; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
+; SSE2-NEXT: pcmpgtd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; SSE2-NEXT: andpd %xmm2, %xmm0
+; SSE2-NEXT: andnpd %xmm1, %xmm2
+; SSE2-NEXT: orpd %xmm2, %xmm0
; SSE2-NEXT: retq
;
; SSE41-LABEL: trunc_signed_v2f64_nsz:
@@ -614,21 +656,44 @@ define <2 x double> @trunc_signed_v2f64_nsz(<2 x double> %x) #0 {
define <4 x double> @trunc_signed_v4f64_nsz(<4 x double> %x) #0 {
; SSE2-LABEL: trunc_signed_v4f64_nsz:
; SSE2: # %bb.0:
-; SSE2-NEXT: cvttsd2si %xmm1, %rax
-; SSE2-NEXT: unpckhpd {{.*#+}} xmm1 = xmm1[1,1]
-; SSE2-NEXT: cvttsd2si %xmm1, %rcx
-; SSE2-NEXT: cvttsd2si %xmm0, %rdx
-; SSE2-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1,1]
-; SSE2-NEXT: cvttsd2si %xmm0, %rsi
-; SSE2-NEXT: xorps %xmm0, %xmm0
-; SSE2-NEXT: cvtsi2sd %rdx, %xmm0
-; SSE2-NEXT: xorps %xmm1, %xmm1
-; SSE2-NEXT: cvtsi2sd %rsi, %xmm1
-; SSE2-NEXT: unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; SSE2-NEXT: xorps %xmm1, %xmm1
-; SSE2-NEXT: cvtsi2sd %rax, %xmm1
-; SSE2-NEXT: cvtsi2sd %rcx, %xmm2
-; SSE2-NEXT: unpcklpd {{.*#+}} xmm1 = xmm1[0],xmm2[0]
+; SSE2-NEXT: movapd {{.*#+}} xmm2 = [NaN,NaN]
+; SSE2-NEXT: movapd %xmm0, %xmm3
+; SSE2-NEXT: andpd %xmm2, %xmm3
+; SSE2-NEXT: cvttsd2si %xmm3, %rax
+; SSE2-NEXT: cvtsi2sd %rax, %xmm4
+; SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm3[2,3,2,3]
+; SSE2-NEXT: cvttsd2si %xmm5, %rax
+; SSE2-NEXT: xorps %xmm5, %xmm5
+; SSE2-NEXT: cvtsi2sd %rax, %xmm5
+; SSE2-NEXT: unpcklpd {{.*#+}} xmm4 = xmm4[0],xmm5[0]
+; SSE2-NEXT: andpd %xmm2, %xmm4
+; SSE2-NEXT: movapd %xmm2, %xmm5
+; SSE2-NEXT: andnpd %xmm0, %xmm5
+; SSE2-NEXT: orpd %xmm4, %xmm5
+; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
+; SSE2-NEXT: movdqa {{.*#+}} xmm4 = [1127219199,1127219199,1127219199,1127219199]
+; SSE2-NEXT: pcmpgtd %xmm4, %xmm3
+; SSE2-NEXT: andpd %xmm3, %xmm0
+; SSE2-NEXT: andnpd %xmm5, %xmm3
+; SSE2-NEXT: orpd %xmm3, %xmm0
+; SSE2-NEXT: movapd %xmm1, %xmm3
+; SSE2-NEXT: andpd %xmm2, %xmm3
+; SSE2-NEXT: cvttsd2si %xmm3, %rax
+; SSE2-NEXT: xorps %xmm5, %xmm5
+; SSE2-NEXT: cvtsi2sd %rax, %xmm5
+; SSE2-NEXT: pshufd {{.*#+}} xmm6 = xmm3[2,3,2,3]
+; SSE2-NEXT: cvttsd2si %xmm6, %rax
+; SSE2-NEXT: xorps %xmm6, %xmm6
+; SSE2-NEXT: cvtsi2sd %rax, %xmm6
+; SSE2-NEXT: unpcklpd {{.*#+}} xmm5 = xmm5[0],xmm6[0]
+; SSE2-NEXT: andpd %xmm2, %xmm5
+; SSE2-NEXT: andnpd %xmm1, %xmm2
+; SSE2-NEXT: orpd %xmm5, %xmm2
+; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
+; SSE2-NEXT: pcmpgtd %xmm4, %xmm3
+; SSE2-NEXT: andpd %xmm3, %xmm1
+; SSE2-NEXT: andnpd %xmm2, %xmm3
+; SSE2-NEXT: orpd %xmm3, %xmm1
; SSE2-NEXT: retq
;
; SSE41-LABEL: trunc_signed_v4f64_nsz:
diff --git a/llvm/test/CodeGen/X86/isel-ftrunc.ll b/llvm/test/CodeGen/X86/isel-ftrunc.ll
index c2be47ca04567..15b98688569bd 100644
--- a/llvm/test/CodeGen/X86/isel-ftrunc.ll
+++ b/llvm/test/CodeGen/X86/isel-ftrunc.ll
@@ -7,16 +7,23 @@
; RUN: llc < %s -mtriple=i686-linux-gnu -global-isel -global-isel-abort=1 | FileCheck %s --check-prefixes=GISEL-X86
define float @trunc_f32(float %a) nounwind readnone {
-; DAG-X64-LABEL: trunc_f32:
-; DAG-X64: # %bb.0:
-; DAG-X64-NEXT: jmp truncf at PLT # TAILCALL
-;
-; FASTISEL-X64-LABEL: trunc_f32:
-; FASTISEL-X64: # %bb.0:
-; FASTISEL-X64-NEXT: pushq %rax
-; FASTISEL-X64-NEXT: callq truncf at PLT
-; FASTISEL-X64-NEXT: popq %rax
-; FASTISEL-X64-NEXT: retq
+; X64-LABEL: trunc_f32:
+; X64: # %bb.0:
+; X64-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN]
+; X64-NEXT: movdqa %xmm0, %xmm2
+; X64-NEXT: pand %xmm1, %xmm2
+; X64-NEXT: movd %xmm2, %eax
+; X64-NEXT: cmpl $1325400064, %eax # imm = 0x4F000000
+; X64-NEXT: jge .LBB0_2
+; X64-NEXT: # %bb.1:
+; X64-NEXT: cvttps2dq %xmm2, %xmm2
+; X64-NEXT: cvtdq2ps %xmm2, %xmm2
+; X64-NEXT: andps %xmm1, %xmm2
+; X64-NEXT: pandn %xmm0, %xmm1
+; X64-NEXT: por %xmm2, %xmm1
+; X64-NEXT: movdqa %xmm1, %xmm0
+; X64-NEXT: .LBB0_2:
+; X64-NEXT: retq
;
; X86-LABEL: trunc_f32:
; X86: # %bb.0:
@@ -47,16 +54,25 @@ define float @trunc_f32(float %a) nounwind readnone {
}
define double @trunc_f64(double %a) nounwind readnone {
-; DAG-X64-LABEL: trunc_f64:
-; DAG-X64: # %bb.0:
-; DAG-X64-NEXT: jmp trunc at PLT # TAILCALL
-;
-; FASTISEL-X64-LABEL: trunc_f64:
-; FASTISEL-X64: # %bb.0:
-; FASTISEL-X64-NEXT: pushq %rax
-; FASTISEL-X64-NEXT: callq trunc at PLT
-; FASTISEL-X64-NEXT: popq %rax
-; FASTISEL-X64-NEXT: retq
+; X64-LABEL: trunc_f64:
+; X64: # %bb.0:
+; X64-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN]
+; X64-NEXT: movdqa %xmm0, %xmm2
+; X64-NEXT: pand %xmm1, %xmm2
+; X64-NEXT: movq %xmm2, %rax
+; X64-NEXT: movabsq $4841369599423283199, %rcx # imm = 0x432FFFFFFFFFFFFF
+; X64-NEXT: cmpq %rcx, %rax
+; X64-NEXT: jg .LBB1_2
+; X64-NEXT: # %bb.1:
+; X64-NEXT: cvttsd2si %xmm2, %rax
+; X64-NEXT: xorps %xmm2, %xmm2
+; X64-NEXT: cvtsi2sd %rax, %xmm2
+; X64-NEXT: andpd %xmm1, %xmm2
+; X64-NEXT: pandn %xmm0, %xmm1
+; X64-NEXT: por %xmm2, %xmm1
+; X64-NEXT: movdqa %xmm1, %xmm0
+; X64-NEXT: .LBB1_2:
+; X64-NEXT: retq
;
; X86-LABEL: trunc_f64:
; X86: # %bb.0:
@@ -128,3 +144,6 @@ define x86_fp80 @trunc_f80(x86_fp80 %a) nounwind readnone {
%c = call x86_fp80 @llvm.trunc.f80(x86_fp80 %a)
ret x86_fp80 %c
}
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; DAG-X64: {{.*}}
+; FASTISEL-X64: {{.*}}
diff --git a/llvm/test/CodeGen/X86/isint.ll b/llvm/test/CodeGen/X86/isint.ll
index 778cfdf9e8351..c72e6c9a2c457 100644
--- a/llvm/test/CodeGen/X86/isint.ll
+++ b/llvm/test/CodeGen/X86/isint.ll
@@ -20,10 +20,25 @@ define i32 @isint_return(double %d) nounwind {
;
; X64-SSE2-LABEL: isint_return:
; X64-SSE2: # %bb.0:
-; X64-SSE2-NEXT: cvttpd2dq %xmm0, %xmm1
-; X64-SSE2-NEXT: cvtdq2pd %xmm1, %xmm1
-; X64-SSE2-NEXT: cmpeqsd %xmm0, %xmm1
-; X64-SSE2-NEXT: movq %xmm1, %rax
+; X64-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN]
+; X64-SSE2-NEXT: movdqa %xmm0, %xmm2
+; X64-SSE2-NEXT: pand %xmm1, %xmm2
+; X64-SSE2-NEXT: movq %xmm2, %rax
+; X64-SSE2-NEXT: movabsq $4841369599423283199, %rcx # imm = 0x432FFFFFFFFFFFFF
+; X64-SSE2-NEXT: movdqa %xmm0, %xmm3
+; X64-SSE2-NEXT: cmpq %rcx, %rax
+; X64-SSE2-NEXT: jg .LBB0_2
+; X64-SSE2-NEXT: # %bb.1:
+; X64-SSE2-NEXT: cvttsd2si %xmm2, %rax
+; X64-SSE2-NEXT: xorps %xmm2, %xmm2
+; X64-SSE2-NEXT: cvtsi2sd %rax, %xmm2
+; X64-SSE2-NEXT: andpd %xmm1, %xmm2
+; X64-SSE2-NEXT: pandn %xmm0, %xmm1
+; X64-SSE2-NEXT: por %xmm2, %xmm1
+; X64-SSE2-NEXT: movdqa %xmm1, %xmm3
+; X64-SSE2-NEXT: .LBB0_2:
+; X64-SSE2-NEXT: cmpeqsd %xmm3, %xmm0
+; X64-SSE2-NEXT: movq %xmm0, %rax
; X64-SSE2-NEXT: andl $1, %eax
; X64-SSE2-NEXT: # kill: def $eax killed $eax killed $rax
; X64-SSE2-NEXT: retq
@@ -62,20 +77,46 @@ define i32 @isint_return(double %d) nounwind {
define i32 @isint_float_return(float %f) nounwind {
; X86-LABEL: isint_float_return:
; X86: # %bb.0:
-; X86-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; X86-NEXT: cvttps2dq %xmm0, %xmm1
-; X86-NEXT: cvtdq2ps %xmm1, %xmm1
-; X86-NEXT: cmpeqss %xmm0, %xmm1
-; X86-NEXT: movd %xmm1, %eax
+; X86-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X86-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN]
+; X86-NEXT: movdqa %xmm0, %xmm2
+; X86-NEXT: pand %xmm1, %xmm2
+; X86-NEXT: movd %xmm2, %eax
+; X86-NEXT: cmpl $1325400064, %eax # imm = 0x4F000000
+; X86-NEXT: movdqa %xmm0, %xmm3
+; X86-NEXT: jge .LBB1_2
+; X86-NEXT: # %bb.1:
+; X86-NEXT: cvttps2dq %xmm2, %xmm2
+; X86-NEXT: cvtdq2ps %xmm2, %xmm2
+; X86-NEXT: andps %xmm1, %xmm2
+; X86-NEXT: pandn %xmm0, %xmm1
+; X86-NEXT: por %xmm2, %xmm1
+; X86-NEXT: movdqa %xmm1, %xmm3
+; X86-NEXT: .LBB1_2:
+; X86-NEXT: cmpeqss %xmm0, %xmm3
+; X86-NEXT: movd %xmm3, %eax
; X86-NEXT: andl $1, %eax
; X86-NEXT: retl
;
; X64-SSE2-LABEL: isint_float_return:
; X64-SSE2: # %bb.0:
-; X64-SSE2-NEXT: cvttps2dq %xmm0, %xmm1
-; X64-SSE2-NEXT: cvtdq2ps %xmm1, %xmm1
-; X64-SSE2-NEXT: cmpeqss %xmm0, %xmm1
-; X64-SSE2-NEXT: movd %xmm1, %eax
+; X64-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN]
+; X64-SSE2-NEXT: movdqa %xmm0, %xmm2
+; X64-SSE2-NEXT: pand %xmm1, %xmm2
+; X64-SSE2-NEXT: movd %xmm2, %eax
+; X64-SSE2-NEXT: movdqa %xmm0, %xmm3
+; X64-SSE2-NEXT: cmpl $1325400064, %eax # imm = 0x4F000000
+; X64-SSE2-NEXT: jge .LBB1_2
+; X64-SSE2-NEXT: # %bb.1:
+; X64-SSE2-NEXT: cvttps2dq %xmm2, %xmm2
+; X64-SSE2-NEXT: cvtdq2ps %xmm2, %xmm2
+; X64-SSE2-NEXT: andps %xmm1, %xmm2
+; X64-SSE2-NEXT: pandn %xmm0, %xmm1
+; X64-SSE2-NEXT: por %xmm2, %xmm1
+; X64-SSE2-NEXT: movdqa %xmm1, %xmm3
+; X64-SSE2-NEXT: .LBB1_2:
+; X64-SSE2-NEXT: cmpeqss %xmm3, %xmm0
+; X64-SSE2-NEXT: movd %xmm0, %eax
; X64-SSE2-NEXT: andl $1, %eax
; X64-SSE2-NEXT: retq
;
@@ -126,16 +167,31 @@ define void @isint_branch(double %d) nounwind {
;
; X64-SSE2-LABEL: isint_branch:
; X64-SSE2: # %bb.0:
-; X64-SSE2-NEXT: cvttpd2dq %xmm0, %xmm1
-; X64-SSE2-NEXT: cvtdq2pd %xmm1, %xmm1
-; X64-SSE2-NEXT: ucomisd %xmm1, %xmm0
-; X64-SSE2-NEXT: jne .LBB2_2
-; X64-SSE2-NEXT: jp .LBB2_2
-; X64-SSE2-NEXT: # %bb.1: # %true
+; X64-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN]
+; X64-SSE2-NEXT: movdqa %xmm0, %xmm2
+; X64-SSE2-NEXT: pand %xmm1, %xmm2
+; X64-SSE2-NEXT: movq %xmm2, %rax
+; X64-SSE2-NEXT: movabsq $4841369599423283199, %rcx # imm = 0x432FFFFFFFFFFFFF
+; X64-SSE2-NEXT: movdqa %xmm0, %xmm3
+; X64-SSE2-NEXT: cmpq %rcx, %rax
+; X64-SSE2-NEXT: jg .LBB2_2
+; X64-SSE2-NEXT: # %bb.1:
+; X64-SSE2-NEXT: cvttsd2si %xmm2, %rax
+; X64-SSE2-NEXT: xorps %xmm2, %xmm2
+; X64-SSE2-NEXT: cvtsi2sd %rax, %xmm2
+; X64-SSE2-NEXT: andpd %xmm1, %xmm2
+; X64-SSE2-NEXT: pandn %xmm0, %xmm1
+; X64-SSE2-NEXT: por %xmm2, %xmm1
+; X64-SSE2-NEXT: movdqa %xmm1, %xmm3
+; X64-SSE2-NEXT: .LBB2_2:
+; X64-SSE2-NEXT: ucomisd %xmm3, %xmm0
+; X64-SSE2-NEXT: jne .LBB2_4
+; X64-SSE2-NEXT: jp .LBB2_4
+; X64-SSE2-NEXT: # %bb.3: # %true
; X64-SSE2-NEXT: pushq %rax
; X64-SSE2-NEXT: callq foo at PLT
; X64-SSE2-NEXT: popq %rax
-; X64-SSE2-NEXT: .LBB2_2: # %false
+; X64-SSE2-NEXT: .LBB2_4: # %false
; X64-SSE2-NEXT: retq
;
; X64-SSE42-LABEL: isint_branch:
diff --git a/llvm/test/CodeGen/X86/llround-conv.ll b/llvm/test/CodeGen/X86/llround-conv.ll
index 2b54c90f76906..1779538c61add 100644
--- a/llvm/test/CodeGen/X86/llround-conv.ll
+++ b/llvm/test/CodeGen/X86/llround-conv.ll
@@ -33,7 +33,21 @@ define i64 @test_llround_f16(half %x) nounwind {
; X64: # %bb.0:
; X64-NEXT: pushq %rax
; X64-NEXT: callq __extendhfsf2 at PLT
-; X64-NEXT: callq roundf at PLT
+; X64-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN]
+; X64-NEXT: movdqa %xmm0, %xmm2
+; X64-NEXT: pand %xmm1, %xmm2
+; X64-NEXT: movd %xmm2, %eax
+; X64-NEXT: cmpl $1325400064, %eax # imm = 0x4F000000
+; X64-NEXT: jge .LBB0_2
+; X64-NEXT: # %bb.1:
+; X64-NEXT: addss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; X64-NEXT: cvttps2dq %xmm2, %xmm2
+; X64-NEXT: cvtdq2ps %xmm2, %xmm2
+; X64-NEXT: andps %xmm1, %xmm2
+; X64-NEXT: pandn %xmm0, %xmm1
+; X64-NEXT: por %xmm2, %xmm1
+; X64-NEXT: movdqa %xmm1, %xmm0
+; X64-NEXT: .LBB0_2:
; X64-NEXT: callq __truncsfhf2 at PLT
; X64-NEXT: callq __extendhfsf2 at PLT
; X64-NEXT: cvttss2si %xmm0, %rax
@@ -213,7 +227,21 @@ define i64 @test_llround_i64_f16(half %x) nounwind {
; X64: # %bb.0:
; X64-NEXT: pushq %rax
; X64-NEXT: callq __extendhfsf2 at PLT
-; X64-NEXT: callq roundf at PLT
+; X64-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN]
+; X64-NEXT: movdqa %xmm0, %xmm2
+; X64-NEXT: pand %xmm1, %xmm2
+; X64-NEXT: movd %xmm2, %eax
+; X64-NEXT: cmpl $1325400064, %eax # imm = 0x4F000000
+; X64-NEXT: jge .LBB4_2
+; X64-NEXT: # %bb.1:
+; X64-NEXT: addss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; X64-NEXT: cvttps2dq %xmm2, %xmm2
+; X64-NEXT: cvtdq2ps %xmm2, %xmm2
+; X64-NEXT: andps %xmm1, %xmm2
+; X64-NEXT: pandn %xmm0, %xmm1
+; X64-NEXT: por %xmm2, %xmm1
+; X64-NEXT: movdqa %xmm1, %xmm0
+; X64-NEXT: .LBB4_2:
; X64-NEXT: callq __truncsfhf2 at PLT
; X64-NEXT: callq __extendhfsf2 at PLT
; X64-NEXT: cvttss2si %xmm0, %rax
diff --git a/llvm/test/CodeGen/X86/lround-conv-i32.ll b/llvm/test/CodeGen/X86/lround-conv-i32.ll
index 02480d087d884..7c15c8318353a 100644
--- a/llvm/test/CodeGen/X86/lround-conv-i32.ll
+++ b/llvm/test/CodeGen/X86/lround-conv-i32.ll
@@ -19,28 +19,57 @@ define i32 @test_lround_i32_f16(half %x) nounwind {
;
; X86-SSE2-LABEL: test_lround_i32_f16:
; X86-SSE2: # %bb.0:
-; X86-SSE2-NEXT: subl $8, %esp
+; X86-SSE2-NEXT: subl $12, %esp
; X86-SSE2-NEXT: pinsrw $0, {{[0-9]+}}(%esp), %xmm0
; X86-SSE2-NEXT: pextrw $0, %xmm0, %eax
; X86-SSE2-NEXT: movw %ax, (%esp)
; X86-SSE2-NEXT: calll __extendhfsf2
-; X86-SSE2-NEXT: fstps (%esp)
-; X86-SSE2-NEXT: calll roundf
-; X86-SSE2-NEXT: fstps (%esp)
+; X86-SSE2-NEXT: fstps {{[0-9]+}}(%esp)
+; X86-SSE2-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; X86-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN]
+; X86-SSE2-NEXT: movdqa %xmm0, %xmm2
+; X86-SSE2-NEXT: pand %xmm1, %xmm2
+; X86-SSE2-NEXT: movd %xmm2, %eax
+; X86-SSE2-NEXT: cmpl $1325400064, %eax # imm = 0x4F000000
+; X86-SSE2-NEXT: jge .LBB0_2
+; X86-SSE2-NEXT: # %bb.1:
+; X86-SSE2-NEXT: addss {{\.?LCPI[0-9]+_[0-9]+}}, %xmm2
+; X86-SSE2-NEXT: cvttps2dq %xmm2, %xmm2
+; X86-SSE2-NEXT: cvtdq2ps %xmm2, %xmm2
+; X86-SSE2-NEXT: andps %xmm1, %xmm2
+; X86-SSE2-NEXT: pandn %xmm0, %xmm1
+; X86-SSE2-NEXT: por %xmm2, %xmm1
+; X86-SSE2-NEXT: movdqa %xmm1, %xmm0
+; X86-SSE2-NEXT: .LBB0_2:
+; X86-SSE2-NEXT: movd %xmm0, (%esp)
; X86-SSE2-NEXT: calll __truncsfhf2
; X86-SSE2-NEXT: pextrw $0, %xmm0, %eax
; X86-SSE2-NEXT: movw %ax, (%esp)
; X86-SSE2-NEXT: calll __extendhfsf2
; X86-SSE2-NEXT: fstps {{[0-9]+}}(%esp)
; X86-SSE2-NEXT: cvttss2si {{[0-9]+}}(%esp), %eax
-; X86-SSE2-NEXT: addl $8, %esp
+; X86-SSE2-NEXT: addl $12, %esp
; X86-SSE2-NEXT: retl
;
; X64-LABEL: test_lround_i32_f16:
; X64: # %bb.0:
; X64-NEXT: pushq %rax
; X64-NEXT: callq __extendhfsf2 at PLT
-; X64-NEXT: callq roundf at PLT
+; X64-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN]
+; X64-NEXT: movdqa %xmm0, %xmm2
+; X64-NEXT: pand %xmm1, %xmm2
+; X64-NEXT: movd %xmm2, %eax
+; X64-NEXT: cmpl $1325400064, %eax # imm = 0x4F000000
+; X64-NEXT: jge .LBB0_2
+; X64-NEXT: # %bb.1:
+; X64-NEXT: addss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; X64-NEXT: cvttps2dq %xmm2, %xmm2
+; X64-NEXT: cvtdq2ps %xmm2, %xmm2
+; X64-NEXT: andps %xmm1, %xmm2
+; X64-NEXT: pandn %xmm0, %xmm1
+; X64-NEXT: por %xmm2, %xmm1
+; X64-NEXT: movdqa %xmm1, %xmm0
+; X64-NEXT: .LBB0_2:
; X64-NEXT: callq __truncsfhf2 at PLT
; X64-NEXT: callq __extendhfsf2 at PLT
; X64-NEXT: cvttss2si %xmm0, %eax
diff --git a/llvm/test/CodeGen/X86/lround-conv-i64.ll b/llvm/test/CodeGen/X86/lround-conv-i64.ll
index f07e4a74bd5e3..fc9aa38a8475b 100644
--- a/llvm/test/CodeGen/X86/lround-conv-i64.ll
+++ b/llvm/test/CodeGen/X86/lround-conv-i64.ll
@@ -33,7 +33,21 @@ define i64 @test_lround_i64_f16(half %x) nounwind {
; X64: # %bb.0: # %entry
; X64-NEXT: pushq %rax
; X64-NEXT: callq __extendhfsf2 at PLT
-; X64-NEXT: callq roundf at PLT
+; X64-NEXT: movdqa {{.*#+}} xmm1 = [NaN,NaN,NaN,NaN]
+; X64-NEXT: movdqa %xmm0, %xmm2
+; X64-NEXT: pand %xmm1, %xmm2
+; X64-NEXT: movd %xmm2, %eax
+; X64-NEXT: cmpl $1325400064, %eax # imm = 0x4F000000
+; X64-NEXT: jge .LBB0_2
+; X64-NEXT: # %bb.1: # %entry
+; X64-NEXT: addss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; X64-NEXT: cvttps2dq %xmm2, %xmm2
+; X64-NEXT: cvtdq2ps %xmm2, %xmm2
+; X64-NEXT: andps %xmm1, %xmm2
+; X64-NEXT: pandn %xmm0, %xmm1
+; X64-NEXT: por %xmm2, %xmm1
+; X64-NEXT: movdqa %xmm1, %xmm0
+; X64-NEXT: .LBB0_2: # %entry
; X64-NEXT: callq __truncsfhf2 at PLT
; X64-NEXT: callq __extendhfsf2 at PLT
; X64-NEXT: cvttss2si %xmm0, %rax
diff --git a/llvm/test/CodeGen/X86/unpredictable-brcond.ll b/llvm/test/CodeGen/X86/unpredictable-brcond.ll
index bb44171d20d65..ebf1178043d20 100644
--- a/llvm/test/CodeGen/X86/unpredictable-brcond.ll
+++ b/llvm/test/CodeGen/X86/unpredictable-brcond.ll
@@ -80,15 +80,34 @@ false:
define void @isint_branch(double %d) nounwind {
; CHECK-LABEL: name: isint_branch
; CHECK: bb.0 (%ir-block.0):
- ; CHECK-NEXT: successors: %bb.1(0x50000000), %bb.2(0x30000000)
+ ; CHECK-NEXT: successors: %bb.3(0x40000000), %bb.4(0x40000000)
; CHECK-NEXT: liveins: $xmm0
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:fr64 = COPY $xmm0
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vr128 = COPY [[COPY]]
- ; CHECK-NEXT: [[CVTTPD2DQrr:%[0-9]+]]:vr128 = nofpexcept CVTTPD2DQrr killed [[COPY1]], implicit $mxcsr
- ; CHECK-NEXT: [[CVTDQ2PDrr:%[0-9]+]]:vr128 = CVTDQ2PDrr killed [[CVTTPD2DQrr]]
- ; CHECK-NEXT: [[COPY2:%[0-9]+]]:fr64 = COPY killed [[CVTDQ2PDrr]]
- ; CHECK-NEXT: nofpexcept UCOMISDrr [[COPY]], killed [[COPY2]], implicit-def $eflags, implicit $mxcsr
+ ; CHECK-NEXT: [[MOVAPDrm:%[0-9]+]]:vr128 = MOVAPDrm $rip, 1, $noreg, %const.0, $noreg :: (load (s128) from constant-pool)
+ ; CHECK-NEXT: [[PANDNrr:%[0-9]+]]:vr128 = PANDNrr [[MOVAPDrm]], [[COPY1]]
+ ; CHECK-NEXT: [[PANDrr:%[0-9]+]]:vr128 = PANDrr [[COPY1]], [[MOVAPDrm]]
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:fr64 = COPY killed [[PANDrr]]
+ ; CHECK-NEXT: [[CVTTSD2SI64rr:%[0-9]+]]:gr64 = nofpexcept CVTTSD2SI64rr [[COPY2]], implicit $mxcsr
+ ; CHECK-NEXT: [[CVTSI642SDrr:%[0-9]+]]:fr64 = nofpexcept CVTSI642SDrr killed [[CVTTSD2SI64rr]], implicit $mxcsr
+ ; CHECK-NEXT: [[COPY3:%[0-9]+]]:vr128 = COPY killed [[CVTSI642SDrr]]
+ ; CHECK-NEXT: [[PANDrr1:%[0-9]+]]:vr128 = PANDrr [[COPY3]], [[MOVAPDrm]]
+ ; CHECK-NEXT: [[PORrr:%[0-9]+]]:vr128 = PORrr [[PANDrr1]], killed [[PANDNrr]]
+ ; CHECK-NEXT: [[COPY4:%[0-9]+]]:fr64 = COPY killed [[PORrr]]
+ ; CHECK-NEXT: [[MOVSDto64rr:%[0-9]+]]:gr64 = MOVSDto64rr [[COPY2]]
+ ; CHECK-NEXT: [[MOV64ri:%[0-9]+]]:gr64 = MOV64ri 4841369599423283199
+ ; CHECK-NEXT: [[SUB64rr:%[0-9]+]]:gr64 = SUB64rr [[MOVSDto64rr]], killed [[MOV64ri]], implicit-def $eflags
+ ; CHECK-NEXT: JCC_1 %bb.4, 15, implicit $eflags
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: bb.3 (%ir-block.0):
+ ; CHECK-NEXT: successors: %bb.4(0x80000000)
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: bb.4 (%ir-block.0):
+ ; CHECK-NEXT: successors: %bb.1(0x50000000), %bb.2(0x30000000)
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[PHI:%[0-9]+]]:fr64 = PHI [[COPY4]], %bb.3, [[COPY]], %bb.0
+ ; CHECK-NEXT: nofpexcept UCOMISDrr [[COPY]], killed [[PHI]], implicit-def $eflags, implicit $mxcsr
; CHECK-NEXT: unpredictable JCC_1 %bb.2, 5, implicit $eflags
; CHECK-NEXT: unpredictable JCC_1 %bb.2, 10, implicit $eflags
; CHECK-NEXT: JMP_1 %bb.1
diff --git a/llvm/test/Transforms/SLPVectorizer/X86/fround.ll b/llvm/test/Transforms/SLPVectorizer/X86/fround.ll
index ac23a7b2e2921..7d0943a34d982 100644
--- a/llvm/test/Transforms/SLPVectorizer/X86/fround.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/fround.ll
@@ -715,12 +715,9 @@ define void @rint_8f64() #0 {
define void @trunc_2f64() #0 {
; SSE2-LABEL: @trunc_2f64(
-; SSE2-NEXT: [[LD0:%.*]] = load double, ptr @src64, align 8
-; SSE2-NEXT: [[LD1:%.*]] = load double, ptr getelementptr inbounds ([8 x double], ptr @src64, i32 0, i64 1), align 8
-; SSE2-NEXT: [[TRUNC0:%.*]] = call double @llvm.trunc.f64(double [[LD0]])
-; SSE2-NEXT: [[TRUNC1:%.*]] = call double @llvm.trunc.f64(double [[LD1]])
-; SSE2-NEXT: store double [[TRUNC0]], ptr @dst64, align 8
-; SSE2-NEXT: store double [[TRUNC1]], ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 1), align 8
+; SSE2-NEXT: [[TMP1:%.*]] = load <2 x double>, ptr @src64, align 8
+; SSE2-NEXT: [[TMP2:%.*]] = call <2 x double> @llvm.trunc.v2f64(<2 x double> [[TMP1]])
+; SSE2-NEXT: store <2 x double> [[TMP2]], ptr @dst64, align 8
; SSE2-NEXT: ret void
;
; SSE41-LABEL: @trunc_2f64(
@@ -746,18 +743,12 @@ define void @trunc_2f64() #0 {
define void @trunc_4f64() #0 {
; SSE2-LABEL: @trunc_4f64(
-; SSE2-NEXT: [[LD0:%.*]] = load double, ptr @src64, align 8
-; SSE2-NEXT: [[LD1:%.*]] = load double, ptr getelementptr inbounds ([8 x double], ptr @src64, i32 0, i64 1), align 8
-; SSE2-NEXT: [[LD2:%.*]] = load double, ptr getelementptr inbounds ([8 x double], ptr @src64, i32 0, i64 2), align 8
-; SSE2-NEXT: [[LD3:%.*]] = load double, ptr getelementptr inbounds ([8 x double], ptr @src64, i32 0, i64 3), align 8
-; SSE2-NEXT: [[TRUNC0:%.*]] = call double @llvm.trunc.f64(double [[LD0]])
-; SSE2-NEXT: [[TRUNC1:%.*]] = call double @llvm.trunc.f64(double [[LD1]])
-; SSE2-NEXT: [[TRUNC2:%.*]] = call double @llvm.trunc.f64(double [[LD2]])
-; SSE2-NEXT: [[TRUNC3:%.*]] = call double @llvm.trunc.f64(double [[LD3]])
-; SSE2-NEXT: store double [[TRUNC0]], ptr @dst64, align 8
-; SSE2-NEXT: store double [[TRUNC1]], ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 1), align 8
-; SSE2-NEXT: store double [[TRUNC2]], ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 2), align 8
-; SSE2-NEXT: store double [[TRUNC3]], ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 3), align 8
+; SSE2-NEXT: [[TMP1:%.*]] = load <2 x double>, ptr @src64, align 8
+; SSE2-NEXT: [[TMP2:%.*]] = call <2 x double> @llvm.trunc.v2f64(<2 x double> [[TMP1]])
+; SSE2-NEXT: store <2 x double> [[TMP2]], ptr @dst64, align 8
+; SSE2-NEXT: [[TMP3:%.*]] = load <2 x double>, ptr getelementptr inbounds ([8 x double], ptr @src64, i32 0, i64 2), align 8
+; SSE2-NEXT: [[TMP4:%.*]] = call <2 x double> @llvm.trunc.v2f64(<2 x double> [[TMP3]])
+; SSE2-NEXT: store <2 x double> [[TMP4]], ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 2), align 8
; SSE2-NEXT: ret void
;
; SSE41-LABEL: @trunc_4f64(
@@ -792,30 +783,18 @@ define void @trunc_4f64() #0 {
define void @trunc_8f64() #0 {
; SSE2-LABEL: @trunc_8f64(
-; SSE2-NEXT: [[LD0:%.*]] = load double, ptr @src64, align 8
-; SSE2-NEXT: [[LD1:%.*]] = load double, ptr getelementptr inbounds ([8 x double], ptr @src64, i32 0, i64 1), align 8
-; SSE2-NEXT: [[LD2:%.*]] = load double, ptr getelementptr inbounds ([8 x double], ptr @src64, i32 0, i64 2), align 8
-; SSE2-NEXT: [[LD3:%.*]] = load double, ptr getelementptr inbounds ([8 x double], ptr @src64, i32 0, i64 3), align 8
-; SSE2-NEXT: [[LD4:%.*]] = load double, ptr getelementptr inbounds ([8 x double], ptr @src64, i32 0, i64 4), align 8
-; SSE2-NEXT: [[LD5:%.*]] = load double, ptr getelementptr inbounds ([8 x double], ptr @src64, i32 0, i64 5), align 8
-; SSE2-NEXT: [[LD6:%.*]] = load double, ptr getelementptr inbounds ([8 x double], ptr @src64, i32 0, i64 6), align 8
-; SSE2-NEXT: [[LD7:%.*]] = load double, ptr getelementptr inbounds ([8 x double], ptr @src64, i32 0, i64 7), align 8
-; SSE2-NEXT: [[TRUNC0:%.*]] = call double @llvm.trunc.f64(double [[LD0]])
-; SSE2-NEXT: [[TRUNC1:%.*]] = call double @llvm.trunc.f64(double [[LD1]])
-; SSE2-NEXT: [[TRUNC2:%.*]] = call double @llvm.trunc.f64(double [[LD2]])
-; SSE2-NEXT: [[TRUNC3:%.*]] = call double @llvm.trunc.f64(double [[LD3]])
-; SSE2-NEXT: [[TRUNC4:%.*]] = call double @llvm.trunc.f64(double [[LD4]])
-; SSE2-NEXT: [[TRUNC5:%.*]] = call double @llvm.trunc.f64(double [[LD5]])
-; SSE2-NEXT: [[TRUNC6:%.*]] = call double @llvm.trunc.f64(double [[LD6]])
-; SSE2-NEXT: [[TRUNC7:%.*]] = call double @llvm.trunc.f64(double [[LD7]])
-; SSE2-NEXT: store double [[TRUNC0]], ptr @dst64, align 8
-; SSE2-NEXT: store double [[TRUNC1]], ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 1), align 8
-; SSE2-NEXT: store double [[TRUNC2]], ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 2), align 8
-; SSE2-NEXT: store double [[TRUNC3]], ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 3), align 8
-; SSE2-NEXT: store double [[TRUNC4]], ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 4), align 8
-; SSE2-NEXT: store double [[TRUNC5]], ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 5), align 8
-; SSE2-NEXT: store double [[TRUNC6]], ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 6), align 8
-; SSE2-NEXT: store double [[TRUNC7]], ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 7), align 8
+; SSE2-NEXT: [[TMP1:%.*]] = load <2 x double>, ptr @src64, align 8
+; SSE2-NEXT: [[TMP2:%.*]] = call <2 x double> @llvm.trunc.v2f64(<2 x double> [[TMP1]])
+; SSE2-NEXT: store <2 x double> [[TMP2]], ptr @dst64, align 8
+; SSE2-NEXT: [[TMP3:%.*]] = load <2 x double>, ptr getelementptr inbounds ([8 x double], ptr @src64, i32 0, i64 2), align 8
+; SSE2-NEXT: [[TMP4:%.*]] = call <2 x double> @llvm.trunc.v2f64(<2 x double> [[TMP3]])
+; SSE2-NEXT: store <2 x double> [[TMP4]], ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 2), align 8
+; SSE2-NEXT: [[TMP5:%.*]] = load <2 x double>, ptr getelementptr inbounds ([8 x double], ptr @src64, i32 0, i64 4), align 8
+; SSE2-NEXT: [[TMP6:%.*]] = call <2 x double> @llvm.trunc.v2f64(<2 x double> [[TMP5]])
+; SSE2-NEXT: store <2 x double> [[TMP6]], ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 4), align 8
+; SSE2-NEXT: [[TMP7:%.*]] = load <2 x double>, ptr getelementptr inbounds ([8 x double], ptr @src64, i32 0, i64 6), align 8
+; SSE2-NEXT: [[TMP8:%.*]] = call <2 x double> @llvm.trunc.v2f64(<2 x double> [[TMP7]])
+; SSE2-NEXT: store <2 x double> [[TMP8]], ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 6), align 8
; SSE2-NEXT: ret void
;
; SSE41-LABEL: @trunc_8f64(
@@ -1906,18 +1885,9 @@ define void @rint_16f32() #0 {
define void @trunc_4f32() #0 {
; SSE2-LABEL: @trunc_4f32(
-; SSE2-NEXT: [[LD0:%.*]] = load float, ptr @src32, align 4
-; SSE2-NEXT: [[LD1:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 1), align 4
-; SSE2-NEXT: [[LD2:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 2), align 4
-; SSE2-NEXT: [[LD3:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 3), align 4
-; SSE2-NEXT: [[TRUNC0:%.*]] = call float @llvm.trunc.f32(float [[LD0]])
-; SSE2-NEXT: [[TRUNC1:%.*]] = call float @llvm.trunc.f32(float [[LD1]])
-; SSE2-NEXT: [[TRUNC2:%.*]] = call float @llvm.trunc.f32(float [[LD2]])
-; SSE2-NEXT: [[TRUNC3:%.*]] = call float @llvm.trunc.f32(float [[LD3]])
-; SSE2-NEXT: store float [[TRUNC0]], ptr @dst32, align 4
-; SSE2-NEXT: store float [[TRUNC1]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 1), align 4
-; SSE2-NEXT: store float [[TRUNC2]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 2), align 4
-; SSE2-NEXT: store float [[TRUNC3]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 3), align 4
+; SSE2-NEXT: [[TMP1:%.*]] = load <4 x float>, ptr @src32, align 4
+; SSE2-NEXT: [[TMP2:%.*]] = call <4 x float> @llvm.trunc.v4f32(<4 x float> [[TMP1]])
+; SSE2-NEXT: store <4 x float> [[TMP2]], ptr @dst32, align 4
; SSE2-NEXT: ret void
;
; SSE41-LABEL: @trunc_4f32(
@@ -1949,30 +1919,12 @@ define void @trunc_4f32() #0 {
define void @trunc_8f32() #0 {
; SSE2-LABEL: @trunc_8f32(
-; SSE2-NEXT: [[LD0:%.*]] = load float, ptr @src32, align 4
-; SSE2-NEXT: [[LD1:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 1), align 4
-; SSE2-NEXT: [[LD2:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 2), align 4
-; SSE2-NEXT: [[LD3:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 3), align 4
-; SSE2-NEXT: [[LD4:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 4), align 4
-; SSE2-NEXT: [[LD5:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 5), align 4
-; SSE2-NEXT: [[LD6:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 6), align 4
-; SSE2-NEXT: [[LD7:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 7), align 4
-; SSE2-NEXT: [[TRUNC0:%.*]] = call float @llvm.trunc.f32(float [[LD0]])
-; SSE2-NEXT: [[TRUNC1:%.*]] = call float @llvm.trunc.f32(float [[LD1]])
-; SSE2-NEXT: [[TRUNC2:%.*]] = call float @llvm.trunc.f32(float [[LD2]])
-; SSE2-NEXT: [[TRUNC3:%.*]] = call float @llvm.trunc.f32(float [[LD3]])
-; SSE2-NEXT: [[TRUNC4:%.*]] = call float @llvm.trunc.f32(float [[LD4]])
-; SSE2-NEXT: [[TRUNC5:%.*]] = call float @llvm.trunc.f32(float [[LD5]])
-; SSE2-NEXT: [[TRUNC6:%.*]] = call float @llvm.trunc.f32(float [[LD6]])
-; SSE2-NEXT: [[TRUNC7:%.*]] = call float @llvm.trunc.f32(float [[LD7]])
-; SSE2-NEXT: store float [[TRUNC0]], ptr @dst32, align 4
-; SSE2-NEXT: store float [[TRUNC1]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 1), align 4
-; SSE2-NEXT: store float [[TRUNC2]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 2), align 4
-; SSE2-NEXT: store float [[TRUNC3]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 3), align 4
-; SSE2-NEXT: store float [[TRUNC4]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 4), align 4
-; SSE2-NEXT: store float [[TRUNC5]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 5), align 4
-; SSE2-NEXT: store float [[TRUNC6]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 6), align 4
-; SSE2-NEXT: store float [[TRUNC7]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 7), align 4
+; SSE2-NEXT: [[TMP1:%.*]] = load <4 x float>, ptr @src32, align 4
+; SSE2-NEXT: [[TMP2:%.*]] = call <4 x float> @llvm.trunc.v4f32(<4 x float> [[TMP1]])
+; SSE2-NEXT: store <4 x float> [[TMP2]], ptr @dst32, align 4
+; SSE2-NEXT: [[TMP3:%.*]] = load <4 x float>, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 4), align 4
+; SSE2-NEXT: [[TMP4:%.*]] = call <4 x float> @llvm.trunc.v4f32(<4 x float> [[TMP3]])
+; SSE2-NEXT: store <4 x float> [[TMP4]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 4), align 4
; SSE2-NEXT: ret void
;
; SSE41-LABEL: @trunc_8f32(
@@ -2019,54 +1971,18 @@ define void @trunc_8f32() #0 {
define void @trunc_16f32() #0 {
; SSE2-LABEL: @trunc_16f32(
-; SSE2-NEXT: [[LD0:%.*]] = load float, ptr @src32, align 4
-; SSE2-NEXT: [[LD1:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 1), align 4
-; SSE2-NEXT: [[LD2:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 2), align 4
-; SSE2-NEXT: [[LD3:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 3), align 4
-; SSE2-NEXT: [[LD4:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 4), align 4
-; SSE2-NEXT: [[LD5:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 5), align 4
-; SSE2-NEXT: [[LD6:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 6), align 4
-; SSE2-NEXT: [[LD7:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 7), align 4
-; SSE2-NEXT: [[LD8:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 8), align 4
-; SSE2-NEXT: [[LD9:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 9), align 4
-; SSE2-NEXT: [[LD10:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 10), align 4
-; SSE2-NEXT: [[LD11:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 11), align 4
-; SSE2-NEXT: [[LD12:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 12), align 4
-; SSE2-NEXT: [[LD13:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 13), align 4
-; SSE2-NEXT: [[LD14:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 14), align 4
-; SSE2-NEXT: [[LD15:%.*]] = load float, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 15), align 4
-; SSE2-NEXT: [[TRUNC0:%.*]] = call float @llvm.trunc.f32(float [[LD0]])
-; SSE2-NEXT: [[TRUNC1:%.*]] = call float @llvm.trunc.f32(float [[LD1]])
-; SSE2-NEXT: [[TRUNC2:%.*]] = call float @llvm.trunc.f32(float [[LD2]])
-; SSE2-NEXT: [[TRUNC3:%.*]] = call float @llvm.trunc.f32(float [[LD3]])
-; SSE2-NEXT: [[TRUNC4:%.*]] = call float @llvm.trunc.f32(float [[LD4]])
-; SSE2-NEXT: [[TRUNC5:%.*]] = call float @llvm.trunc.f32(float [[LD5]])
-; SSE2-NEXT: [[TRUNC6:%.*]] = call float @llvm.trunc.f32(float [[LD6]])
-; SSE2-NEXT: [[TRUNC7:%.*]] = call float @llvm.trunc.f32(float [[LD7]])
-; SSE2-NEXT: [[TRUNC8:%.*]] = call float @llvm.trunc.f32(float [[LD8]])
-; SSE2-NEXT: [[TRUNC9:%.*]] = call float @llvm.trunc.f32(float [[LD9]])
-; SSE2-NEXT: [[TRUNC10:%.*]] = call float @llvm.trunc.f32(float [[LD10]])
-; SSE2-NEXT: [[TRUNC11:%.*]] = call float @llvm.trunc.f32(float [[LD11]])
-; SSE2-NEXT: [[TRUNC12:%.*]] = call float @llvm.trunc.f32(float [[LD12]])
-; SSE2-NEXT: [[TRUNC13:%.*]] = call float @llvm.trunc.f32(float [[LD13]])
-; SSE2-NEXT: [[TRUNC14:%.*]] = call float @llvm.trunc.f32(float [[LD14]])
-; SSE2-NEXT: [[TRUNC15:%.*]] = call float @llvm.trunc.f32(float [[LD15]])
-; SSE2-NEXT: store float [[TRUNC0]], ptr @dst32, align 4
-; SSE2-NEXT: store float [[TRUNC1]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 1), align 4
-; SSE2-NEXT: store float [[TRUNC2]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 2), align 4
-; SSE2-NEXT: store float [[TRUNC3]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 3), align 4
-; SSE2-NEXT: store float [[TRUNC4]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 4), align 4
-; SSE2-NEXT: store float [[TRUNC5]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 5), align 4
-; SSE2-NEXT: store float [[TRUNC6]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 6), align 4
-; SSE2-NEXT: store float [[TRUNC7]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 7), align 4
-; SSE2-NEXT: store float [[TRUNC8]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 8), align 4
-; SSE2-NEXT: store float [[TRUNC9]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 9), align 4
-; SSE2-NEXT: store float [[TRUNC10]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 10), align 4
-; SSE2-NEXT: store float [[TRUNC11]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 11), align 4
-; SSE2-NEXT: store float [[TRUNC12]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 12), align 4
-; SSE2-NEXT: store float [[TRUNC13]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 13), align 4
-; SSE2-NEXT: store float [[TRUNC14]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 14), align 4
-; SSE2-NEXT: store float [[TRUNC15]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 15), align 4
+; SSE2-NEXT: [[TMP1:%.*]] = load <4 x float>, ptr @src32, align 4
+; SSE2-NEXT: [[TMP2:%.*]] = call <4 x float> @llvm.trunc.v4f32(<4 x float> [[TMP1]])
+; SSE2-NEXT: store <4 x float> [[TMP2]], ptr @dst32, align 4
+; SSE2-NEXT: [[TMP3:%.*]] = load <4 x float>, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 4), align 4
+; SSE2-NEXT: [[TMP4:%.*]] = call <4 x float> @llvm.trunc.v4f32(<4 x float> [[TMP3]])
+; SSE2-NEXT: store <4 x float> [[TMP4]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 4), align 4
+; SSE2-NEXT: [[TMP5:%.*]] = load <4 x float>, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 8), align 4
+; SSE2-NEXT: [[TMP6:%.*]] = call <4 x float> @llvm.trunc.v4f32(<4 x float> [[TMP5]])
+; SSE2-NEXT: store <4 x float> [[TMP6]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 8), align 4
+; SSE2-NEXT: [[TMP7:%.*]] = load <4 x float>, ptr getelementptr inbounds ([16 x float], ptr @src32, i32 0, i64 12), align 4
+; SSE2-NEXT: [[TMP8:%.*]] = call <4 x float> @llvm.trunc.v4f32(<4 x float> [[TMP7]])
+; SSE2-NEXT: store <4 x float> [[TMP8]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 12), align 4
; SSE2-NEXT: ret void
;
; SSE41-LABEL: @trunc_16f32(
>From 8fe93a8a15d3cd8cc84d8df253d3b708f99712e1 Mon Sep 17 00:00:00 2001
From: Divyansh <anshmcs at gmail.com>
Date: Fri, 25 Sep 2026 21:16:06 +0530
Subject: [PATCH 2/2] Apply clang-format fixes
---
llvm/lib/Target/X86/X86ISelLowering.cpp | 25 ++++++++++++++++---------
1 file changed, 16 insertions(+), 9 deletions(-)
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index c819efa95a515..08d1f549c2c5d 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -1101,7 +1101,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
for (auto VT : { MVT::f64, MVT::v4f32, MVT::v2f64 })
SetFPMinMaxAction(VT);
- for (auto VT : { MVT::v4f32, MVT::v2f64 }) {
+ for (auto VT : {MVT::v4f32, MVT::v2f64}) {
if (VT == MVT::v4f32 || Subtarget.is64Bit()) {
setOperationAction(ISD::FTRUNC, VT, Custom);
setOperationAction(ISD::FROUND, VT, Custom);
@@ -23203,16 +23203,22 @@ static SDValue lowerFTRUNC_FROUND_SSE2(SDValue Op, SelectionDAG &DAG) {
APFloat Bias = APFloat(0.5f);
bool Ignored;
Bias.convert(Sem, APFloat::rmNearestTiesToEven, &Ignored);
- Bias.next(/*nextDown*/true);
- AbsBiased = DAG.getNode(ISD::FADD, DL, VT, Abs, DAG.getConstantFP(Bias, DL, VT));
+ Bias.next(/*nextDown*/ true);
+ AbsBiased =
+ DAG.getNode(ISD::FADD, DL, VT, Abs, DAG.getConstantFP(Bias, DL, VT));
}
MVT IntVT;
- if (VT == MVT::f32) IntVT = MVT::i32;
- else if (VT == MVT::f64) IntVT = MVT::i64;
- else if (VT == MVT::v4f32) IntVT = MVT::v4i32;
- else if (VT == MVT::v2f64) IntVT = MVT::v2i64;
- else llvm_unreachable("Unexpected type");
+ if (VT == MVT::f32)
+ IntVT = MVT::i32;
+ else if (VT == MVT::f64)
+ IntVT = MVT::i64;
+ else if (VT == MVT::v4f32)
+ IntVT = MVT::v4i32;
+ else if (VT == MVT::v2f64)
+ IntVT = MVT::v2i64;
+ else
+ llvm_unreachable("Unexpected type");
SDValue AbsInt = DAG.getBitcast(IntVT, Abs);
@@ -23223,7 +23229,8 @@ static SDValue lowerFTRUNC_FROUND_SSE2(SDValue Op, SelectionDAG &DAG) {
Threshold = DAG.getConstant(0x432FFFFFFFFFFFFFULL, DL, IntVT);
}
- EVT CCVT = DAG.getTargetLoweringInfo().getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(), IntVT);
+ EVT CCVT = DAG.getTargetLoweringInfo().getSetCCResultType(
+ DAG.getDataLayout(), *DAG.getContext(), IntVT);
SDValue IsLarge = DAG.getSetCC(DL, CCVT, AbsInt, Threshold, ISD::SETGT);
SDValue TruncInt = DAG.getNode(ISD::FP_TO_SINT, DL, IntVT, AbsBiased);
More information about the llvm-commits
mailing list