[llvm] [AMDGPU] Move `AMDGPUAttributorPass` to full LTO post link stage (PR #102086)
Shilei Tian via llvm-commits
llvm-commits at lists.llvm.org
Thu Aug 8 16:57:39 PDT 2024
https://github.com/shiltian updated https://github.com/llvm/llvm-project/pull/102086
>From 04478b8899691097228011cdc93044640e666a4a Mon Sep 17 00:00:00 2001
From: Shilei Tian <i at tianshilei.me>
Date: Mon, 5 Aug 2024 20:27:47 -0400
Subject: [PATCH] [AMDGPU] Add an option to run `AMDGPUAttributorPass` in full
LTO
Currently `AMDGPUAttributorPass` is registered in default optimizer pipeline.
This will allow the pass to run in default pipeline as well as at thinLTO post
link stage. However, it will not run in full LTO post link stage. This patch
adds an option to run it in full LTO. It is set to false by default.
---
.../lib/Target/AMDGPU/AMDGPUTargetMachine.cpp | 15 +-
.../AMDGPU/amdgpu-simplify-libcall-sincos.ll | 477 +++++++++---------
llvm/test/CodeGen/AMDGPU/attributor-noopt.ll | 2 +-
llvm/test/CodeGen/AMDGPU/inline-attr.ll | 6 +-
llvm/test/CodeGen/AMDGPU/simplify-libcalls.ll | 2 +-
5 files changed, 242 insertions(+), 260 deletions(-)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
index c8fb68d1c0b0c..c4788d87abae4 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -301,6 +301,11 @@ static cl::opt<bool> EnableSIModeRegisterPass(
cl::init(true),
cl::Hidden);
+static cl::opt<bool>
+ EnableAMDGPUAttributor("amdgpu-attributor-enable",
+ cl::desc("Enable AMDGPUAttributorPass"),
+ cl::init(true), cl::Hidden);
+
// Enable GFX11.5+ s_singleuse_vdst insertion
static cl::opt<bool>
EnableInsertSingleUseVDST("amdgpu-enable-single-use-vdst",
@@ -734,14 +739,6 @@ void AMDGPUTargetMachine::registerPassBuilderCallbacks(PassBuilder &PB) {
PM.addPass(createCGSCCToFunctionPassAdaptor(std::move(FPM)));
});
- // FIXME: Why is AMDGPUAttributor not in CGSCC?
- PB.registerOptimizerLastEPCallback(
- [this](ModulePassManager &MPM, OptimizationLevel Level) {
- if (Level != OptimizationLevel::O0) {
- MPM.addPass(AMDGPUAttributorPass(*this));
- }
- });
-
PB.registerFullLinkTimeOptimizationLastEPCallback(
[this](ModulePassManager &PM, OptimizationLevel Level) {
// We want to support the -lto-partitions=N option as "best effort".
@@ -749,6 +746,8 @@ void AMDGPUTargetMachine::registerPassBuilderCallbacks(PassBuilder &PB) {
// module is partitioned for codegen.
if (EnableLowerModuleLDS)
PM.addPass(AMDGPULowerModuleLDSPass(*this));
+ if (EnableAMDGPUAttributor)
+ PM.addPass(AMDGPUAttributorPass(*this));
});
PB.registerRegClassFilterParsingCallback(
diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-sincos.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-sincos.ll
index 1358d91ae102c..de03dc4a9cd4c 100644
--- a/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-sincos.ll
+++ b/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-sincos.ll
@@ -1,5 +1,5 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
-; RUN: opt -S -O1 -mtriple=amdgcn-- -amdgpu-simplify-libcall=1 -amdgpu-prelink < %s | FileCheck %s
+; RUN: opt -S -passes="lto<O1>" -mtriple=amdgcn-- -amdgpu-simplify-libcall=1 -amdgpu-prelink < %s | FileCheck %s
target datalayout = "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-p7:160:256:256:32-p8:128:128-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5-G1-ni:7:8:9"
@@ -49,7 +49,7 @@ declare float @_Z6sincosfPU3AS0f(float %x, ptr writeonly %ptr) #1
define void @sincos_f16_nocontract(half %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_f16_nocontract
-; CHECK-SAME: (half [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR2:[0-9]+]] {
+; CHECK-SAME: (half [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2:[0-9]+]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CALL:%.*]] = tail call half @_Z3sinDh(half [[X]])
; CHECK-NEXT: store half [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 2
@@ -68,7 +68,7 @@ entry:
define void @sincos_v2f16_nocontract(<2 x half> %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_v2f16_nocontract
-; CHECK-SAME: (<2 x half> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR2]] {
+; CHECK-SAME: (<2 x half> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CALL:%.*]] = tail call <2 x half> @_Z3sinDv2_Dh(<2 x half> [[X]])
; CHECK-NEXT: store <2 x half> [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 4
@@ -87,7 +87,7 @@ entry:
define void @sincos_f16(half %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_f16
-; CHECK-SAME: (half [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR2]] {
+; CHECK-SAME: (half [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CALL:%.*]] = tail call contract half @_Z3sinDh(half [[X]])
; CHECK-NEXT: store half [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 2
@@ -105,7 +105,7 @@ entry:
define void @sincos_f16_order1(half %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_f16_order1
-; CHECK-SAME: (half [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR2]] {
+; CHECK-SAME: (half [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CALL1:%.*]] = tail call contract half @_Z3cosDh(half [[X]])
; CHECK-NEXT: store half [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 2
@@ -123,7 +123,7 @@ entry:
define void @sincos_v2f16(<2 x half> %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_v2f16
-; CHECK-SAME: (<2 x half> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR2]] {
+; CHECK-SAME: (<2 x half> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CALL:%.*]] = tail call contract <2 x half> @_Z3sinDv2_Dh(<2 x half> [[X]])
; CHECK-NEXT: store <2 x half> [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 4
@@ -141,7 +141,7 @@ entry:
define void @sincos_v3f16(<3 x half> %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_v3f16
-; CHECK-SAME: (<3 x half> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR2]] {
+; CHECK-SAME: (<3 x half> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CALL:%.*]] = tail call contract <3 x half> @_Z3sinDv3_Dh(<3 x half> [[X]])
; CHECK-NEXT: [[EXTRACTVEC2:%.*]] = shufflevector <3 x half> [[CALL]], <3 x half> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
@@ -164,7 +164,7 @@ entry:
define void @sincos_v4f16(<4 x half> %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_v4f16
-; CHECK-SAME: (<4 x half> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR2]] {
+; CHECK-SAME: (<4 x half> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CALL:%.*]] = tail call contract <4 x half> @_Z3sinDv4_Dh(<4 x half> [[X]])
; CHECK-NEXT: store <4 x half> [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 8
@@ -182,7 +182,7 @@ entry:
define void @sincos_v8f16(<8 x half> %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_v8f16
-; CHECK-SAME: (<8 x half> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR2]] {
+; CHECK-SAME: (<8 x half> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CALL:%.*]] = tail call contract <8 x half> @_Z3sinDv8_Dh(<8 x half> [[X]])
; CHECK-NEXT: store <8 x half> [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 16
@@ -201,7 +201,7 @@ entry:
define void @sincos_v16f16(<16 x half> %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_v16f16
-; CHECK-SAME: (<16 x half> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR2]] {
+; CHECK-SAME: (<16 x half> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CALL:%.*]] = tail call contract <16 x half> @_Z3sinDv16_Dh(<16 x half> [[X]])
; CHECK-NEXT: store <16 x half> [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 32
@@ -220,13 +220,12 @@ entry:
define void @sincos_f32_nocontract(float %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_f32_nocontract
-; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR3:[0-9]+]] {
+; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca float, align 4, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call float @_Z6sincosfPU3AS5f(float [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(5) [[__SINCOS_]], align 4
-; CHECK-NEXT: store float [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 4
-; CHECK-NEXT: store float [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT: [[CALL:%.*]] = tail call float @_Z3sinf(float [[X]])
+; CHECK-NEXT: store float [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 4
+; CHECK-NEXT: [[CALL1:%.*]] = tail call float @_Z3cosf(float [[X]])
+; CHECK-NEXT: store float [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 4
; CHECK-NEXT: ret void
;
entry:
@@ -240,13 +239,12 @@ entry:
define void @sincos_v2f32_nocontract(<2 x float> %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_v2f32_nocontract
-; CHECK-SAME: (<2 x float> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR4:[0-9]+]] {
+; CHECK-SAME: (<2 x float> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca <2 x float>, align 8, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call <2 x float> @_Z6sincosDv2_fPU3AS5S_(<2 x float> [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load <2 x float>, ptr addrspace(5) [[__SINCOS_]], align 8
-; CHECK-NEXT: store <2 x float> [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 8
-; CHECK-NEXT: store <2 x float> [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 8
+; CHECK-NEXT: [[CALL:%.*]] = tail call <2 x float> @_Z3sinDv2_f(<2 x float> [[X]])
+; CHECK-NEXT: store <2 x float> [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 8
+; CHECK-NEXT: [[CALL1:%.*]] = tail call <2 x float> @_Z3cosDv2_f(<2 x float> [[X]])
+; CHECK-NEXT: store <2 x float> [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 8
; CHECK-NEXT: ret void
;
entry:
@@ -260,13 +258,12 @@ entry:
define void @sincos_f32(float %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_f32
-; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR3]] {
+; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca float, align 4, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract float @_Z6sincosfPU3AS5f(float [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(5) [[__SINCOS_]], align 4
-; CHECK-NEXT: store float [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 4
-; CHECK-NEXT: store float [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract float @_Z3sinf(float [[X]])
+; CHECK-NEXT: store float [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 4
+; CHECK-NEXT: [[CALL1:%.*]] = tail call contract float @_Z3cosf(float [[X]])
+; CHECK-NEXT: store float [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 4
; CHECK-NEXT: ret void
;
entry:
@@ -279,13 +276,12 @@ entry:
define void @sincos_f32_order1(float %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_f32_order1
-; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR3]] {
+; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca float, align 4, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract float @_Z6sincosfPU3AS5f(float [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(5) [[__SINCOS_]], align 4
-; CHECK-NEXT: store float [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 4
-; CHECK-NEXT: store float [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 4
+; CHECK-NEXT: [[CALL1:%.*]] = tail call contract float @_Z3cosf(float [[X]])
+; CHECK-NEXT: store float [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract float @_Z3sinf(float [[X]])
+; CHECK-NEXT: store float [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 4
; CHECK-NEXT: ret void
;
entry:
@@ -298,13 +294,12 @@ entry:
define void @sincos_v2f32(<2 x float> %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_v2f32
-; CHECK-SAME: (<2 x float> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR4]] {
+; CHECK-SAME: (<2 x float> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca <2 x float>, align 8, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract <2 x float> @_Z6sincosDv2_fPU3AS5S_(<2 x float> [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load <2 x float>, ptr addrspace(5) [[__SINCOS_]], align 8
-; CHECK-NEXT: store <2 x float> [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 8
-; CHECK-NEXT: store <2 x float> [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 8
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract <2 x float> @_Z3sinDv2_f(<2 x float> [[X]])
+; CHECK-NEXT: store <2 x float> [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 8
+; CHECK-NEXT: [[CALL1:%.*]] = tail call contract <2 x float> @_Z3cosDv2_f(<2 x float> [[X]])
+; CHECK-NEXT: store <2 x float> [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 8
; CHECK-NEXT: ret void
;
entry:
@@ -317,14 +312,13 @@ entry:
define void @sincos_v3f32(<3 x float> %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_v3f32
-; CHECK-SAME: (<3 x float> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR4]] {
+; CHECK-SAME: (<3 x float> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca <3 x float>, align 16, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract <3 x float> @_Z6sincosDv3_fPU3AS5S_(<3 x float> [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load <3 x float>, ptr addrspace(5) [[__SINCOS_]], align 16
-; CHECK-NEXT: [[EXTRACTVEC2:%.*]] = shufflevector <3 x float> [[TMP0]], <3 x float> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract <3 x float> @_Z3sinDv3_f(<3 x float> [[X]])
+; CHECK-NEXT: [[EXTRACTVEC2:%.*]] = shufflevector <3 x float> [[CALL]], <3 x float> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
; CHECK-NEXT: store <4 x float> [[EXTRACTVEC2]], ptr addrspace(1) [[SIN_OUT]], align 16
-; CHECK-NEXT: [[EXTRACTVEC6:%.*]] = shufflevector <3 x float> [[TMP1]], <3 x float> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
+; CHECK-NEXT: [[CALL5:%.*]] = tail call contract <3 x float> @_Z3cosDv3_f(<3 x float> [[X]])
+; CHECK-NEXT: [[EXTRACTVEC6:%.*]] = shufflevector <3 x float> [[CALL5]], <3 x float> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
; CHECK-NEXT: store <4 x float> [[EXTRACTVEC6]], ptr addrspace(1) [[COS_OUT]], align 16
; CHECK-NEXT: ret void
;
@@ -340,13 +334,12 @@ entry:
define void @sincos_v4f32(<4 x float> %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_v4f32
-; CHECK-SAME: (<4 x float> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR4]] {
+; CHECK-SAME: (<4 x float> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca <4 x float>, align 16, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract <4 x float> @_Z6sincosDv4_fPU3AS5S_(<4 x float> [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load <4 x float>, ptr addrspace(5) [[__SINCOS_]], align 16
-; CHECK-NEXT: store <4 x float> [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 16
-; CHECK-NEXT: store <4 x float> [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 16
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract <4 x float> @_Z3sinDv4_f(<4 x float> [[X]])
+; CHECK-NEXT: store <4 x float> [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 16
+; CHECK-NEXT: [[CALL1:%.*]] = tail call contract <4 x float> @_Z3cosDv4_f(<4 x float> [[X]])
+; CHECK-NEXT: store <4 x float> [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 16
; CHECK-NEXT: ret void
;
entry:
@@ -359,13 +352,12 @@ entry:
define void @sincos_v8f32(<8 x float> %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_v8f32
-; CHECK-SAME: (<8 x float> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR4]] {
+; CHECK-SAME: (<8 x float> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca <8 x float>, align 32, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract <8 x float> @_Z6sincosDv8_fPU3AS5S_(<8 x float> [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load <8 x float>, ptr addrspace(5) [[__SINCOS_]], align 32
-; CHECK-NEXT: store <8 x float> [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 32
-; CHECK-NEXT: store <8 x float> [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 32
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract <8 x float> @_Z3sinDv8_f(<8 x float> [[X]])
+; CHECK-NEXT: store <8 x float> [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 32
+; CHECK-NEXT: [[CALL1:%.*]] = tail call contract <8 x float> @_Z3cosDv8_f(<8 x float> [[X]])
+; CHECK-NEXT: store <8 x float> [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 32
; CHECK-NEXT: ret void
;
entry:
@@ -378,13 +370,12 @@ entry:
define void @sincos_v16f32(<16 x float> %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_v16f32
-; CHECK-SAME: (<16 x float> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR4]] {
+; CHECK-SAME: (<16 x float> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca <16 x float>, align 64, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract <16 x float> @_Z6sincosDv16_fPU3AS5S_(<16 x float> [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load <16 x float>, ptr addrspace(5) [[__SINCOS_]], align 64
-; CHECK-NEXT: store <16 x float> [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 64
-; CHECK-NEXT: store <16 x float> [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 64
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract <16 x float> @_Z3sinDv16_f(<16 x float> [[X]])
+; CHECK-NEXT: store <16 x float> [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 64
+; CHECK-NEXT: [[CALL1:%.*]] = tail call contract <16 x float> @_Z3cosDv16_f(<16 x float> [[X]])
+; CHECK-NEXT: store <16 x float> [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 64
; CHECK-NEXT: ret void
;
entry:
@@ -397,13 +388,12 @@ entry:
define void @sincos_f64_nocontract(double %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_f64_nocontract
-; CHECK-SAME: (double [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR4]] {
+; CHECK-SAME: (double [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca double, align 8, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call double @_Z6sincosdPU3AS5d(double [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(5) [[__SINCOS_]], align 8
-; CHECK-NEXT: store double [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 8
-; CHECK-NEXT: store double [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 8
+; CHECK-NEXT: [[CALL:%.*]] = tail call double @_Z3sind(double [[X]])
+; CHECK-NEXT: store double [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 8
+; CHECK-NEXT: [[CALL1:%.*]] = tail call double @_Z3cosd(double [[X]])
+; CHECK-NEXT: store double [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 8
; CHECK-NEXT: ret void
;
entry:
@@ -417,13 +407,12 @@ entry:
define void @sincos_v2f64_nocontract(<2 x double> %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_v2f64_nocontract
-; CHECK-SAME: (<2 x double> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR4]] {
+; CHECK-SAME: (<2 x double> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca <2 x double>, align 16, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call <2 x double> @_Z6sincosDv2_dPU3AS5S_(<2 x double> [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load <2 x double>, ptr addrspace(5) [[__SINCOS_]], align 16
-; CHECK-NEXT: store <2 x double> [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 16
-; CHECK-NEXT: store <2 x double> [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 16
+; CHECK-NEXT: [[CALL:%.*]] = tail call <2 x double> @_Z3sinDv2_d(<2 x double> [[X]])
+; CHECK-NEXT: store <2 x double> [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 16
+; CHECK-NEXT: [[CALL1:%.*]] = tail call <2 x double> @_Z3cosDv2_d(<2 x double> [[X]])
+; CHECK-NEXT: store <2 x double> [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 16
; CHECK-NEXT: ret void
;
entry:
@@ -436,13 +425,12 @@ entry:
define void @sincos_f64(double %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_f64
-; CHECK-SAME: (double [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR4]] {
+; CHECK-SAME: (double [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca double, align 8, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract double @_Z6sincosdPU3AS5d(double [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(5) [[__SINCOS_]], align 8
-; CHECK-NEXT: store double [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 8
-; CHECK-NEXT: store double [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 8
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract double @_Z3sind(double [[X]])
+; CHECK-NEXT: store double [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 8
+; CHECK-NEXT: [[CALL1:%.*]] = tail call contract double @_Z3cosd(double [[X]])
+; CHECK-NEXT: store double [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 8
; CHECK-NEXT: ret void
;
entry:
@@ -455,13 +443,12 @@ entry:
define void @sincos_f64_order1(double %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_f64_order1
-; CHECK-SAME: (double [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR4]] {
+; CHECK-SAME: (double [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca double, align 8, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract double @_Z6sincosdPU3AS5d(double [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load double, ptr addrspace(5) [[__SINCOS_]], align 8
-; CHECK-NEXT: store double [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 8
-; CHECK-NEXT: store double [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 8
+; CHECK-NEXT: [[CALL1:%.*]] = tail call contract double @_Z3cosd(double [[X]])
+; CHECK-NEXT: store double [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 8
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract double @_Z3sind(double [[X]])
+; CHECK-NEXT: store double [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 8
; CHECK-NEXT: ret void
;
entry:
@@ -474,13 +461,12 @@ entry:
define void @sincos_v2f64(<2 x double> %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_v2f64
-; CHECK-SAME: (<2 x double> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR4]] {
+; CHECK-SAME: (<2 x double> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca <2 x double>, align 16, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract <2 x double> @_Z6sincosDv2_dPU3AS5S_(<2 x double> [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load <2 x double>, ptr addrspace(5) [[__SINCOS_]], align 16
-; CHECK-NEXT: store <2 x double> [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 16
-; CHECK-NEXT: store <2 x double> [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 16
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract <2 x double> @_Z3sinDv2_d(<2 x double> [[X]])
+; CHECK-NEXT: store <2 x double> [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 16
+; CHECK-NEXT: [[CALL1:%.*]] = tail call contract <2 x double> @_Z3cosDv2_d(<2 x double> [[X]])
+; CHECK-NEXT: store <2 x double> [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 16
; CHECK-NEXT: ret void
;
entry:
@@ -493,14 +479,13 @@ entry:
define void @sincos_v3f64(<3 x double> %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_v3f64
-; CHECK-SAME: (<3 x double> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR4]] {
+; CHECK-SAME: (<3 x double> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca <3 x double>, align 32, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract <3 x double> @_Z6sincosDv3_dPU3AS5S_(<3 x double> [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load <3 x double>, ptr addrspace(5) [[__SINCOS_]], align 32
-; CHECK-NEXT: [[EXTRACTVEC2:%.*]] = shufflevector <3 x double> [[TMP0]], <3 x double> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract <3 x double> @_Z3sinDv3_d(<3 x double> [[X]])
+; CHECK-NEXT: [[EXTRACTVEC2:%.*]] = shufflevector <3 x double> [[CALL]], <3 x double> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
; CHECK-NEXT: store <4 x double> [[EXTRACTVEC2]], ptr addrspace(1) [[SIN_OUT]], align 32
-; CHECK-NEXT: [[EXTRACTVEC6:%.*]] = shufflevector <3 x double> [[TMP1]], <3 x double> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
+; CHECK-NEXT: [[CALL5:%.*]] = tail call contract <3 x double> @_Z3cosDv3_d(<3 x double> [[X]])
+; CHECK-NEXT: [[EXTRACTVEC6:%.*]] = shufflevector <3 x double> [[CALL5]], <3 x double> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
; CHECK-NEXT: store <4 x double> [[EXTRACTVEC6]], ptr addrspace(1) [[COS_OUT]], align 32
; CHECK-NEXT: ret void
;
@@ -516,13 +501,12 @@ entry:
define void @sincos_v4f64(<4 x double> %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_v4f64
-; CHECK-SAME: (<4 x double> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR4]] {
+; CHECK-SAME: (<4 x double> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca <4 x double>, align 32, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract <4 x double> @_Z6sincosDv4_dPU3AS5S_(<4 x double> [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load <4 x double>, ptr addrspace(5) [[__SINCOS_]], align 32
-; CHECK-NEXT: store <4 x double> [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 32
-; CHECK-NEXT: store <4 x double> [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 32
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract <4 x double> @_Z3sinDv4_d(<4 x double> [[X]])
+; CHECK-NEXT: store <4 x double> [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 32
+; CHECK-NEXT: [[CALL1:%.*]] = tail call contract <4 x double> @_Z3cosDv4_d(<4 x double> [[X]])
+; CHECK-NEXT: store <4 x double> [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 32
; CHECK-NEXT: ret void
;
entry:
@@ -535,13 +519,12 @@ entry:
define void @sincos_v8f64(<8 x double> %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_v8f64
-; CHECK-SAME: (<8 x double> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR4]] {
+; CHECK-SAME: (<8 x double> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca <8 x double>, align 64, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract <8 x double> @_Z6sincosDv8_dPU3AS5S_(<8 x double> [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load <8 x double>, ptr addrspace(5) [[__SINCOS_]], align 64
-; CHECK-NEXT: store <8 x double> [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 64
-; CHECK-NEXT: store <8 x double> [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 64
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract <8 x double> @_Z3sinDv8_d(<8 x double> [[X]])
+; CHECK-NEXT: store <8 x double> [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 64
+; CHECK-NEXT: [[CALL1:%.*]] = tail call contract <8 x double> @_Z3cosDv8_d(<8 x double> [[X]])
+; CHECK-NEXT: store <8 x double> [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 64
; CHECK-NEXT: ret void
;
entry:
@@ -554,13 +537,12 @@ entry:
define void @sincos_v16f64(<16 x double> %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_v16f64
-; CHECK-SAME: (<16 x double> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR4]] {
+; CHECK-SAME: (<16 x double> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca <16 x double>, align 128, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract <16 x double> @_Z6sincosDv16_dPU3AS5S_(<16 x double> [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load <16 x double>, ptr addrspace(5) [[__SINCOS_]], align 128
-; CHECK-NEXT: store <16 x double> [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 128
-; CHECK-NEXT: store <16 x double> [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 128
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract <16 x double> @_Z3sinDv16_d(<16 x double> [[X]])
+; CHECK-NEXT: store <16 x double> [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 128
+; CHECK-NEXT: [[CALL1:%.*]] = tail call contract <16 x double> @_Z3cosDv16_d(<16 x double> [[X]])
+; CHECK-NEXT: store <16 x double> [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 128
; CHECK-NEXT: ret void
;
entry:
@@ -575,20 +557,17 @@ entry:
define void @sincos_f32_different_blocks(i1 %cond, float %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_f32_different_blocks
-; CHECK-SAME: (i1 [[COND:%.*]], float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR3]] {
+; CHECK-SAME: (i1 [[COND:%.*]], float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca float, align 4, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract float @_Z6sincosfPU3AS5f(float [[X]], ptr addrspace(5) [[__SINCOS_]])
; CHECK-NEXT: br i1 [[COND]], label [[BB0:%.*]], label [[BB1:%.*]]
-; CHECK: common.ret:
-; CHECK-NEXT: ret void
; CHECK: bb0:
-; CHECK-NEXT: store float [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 4
-; CHECK-NEXT: br label [[COMMON_RET:%.*]]
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract float @_Z3sinf(float [[X]])
+; CHECK-NEXT: store float [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 4
+; CHECK-NEXT: ret void
; CHECK: bb1:
-; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(5) [[__SINCOS_]], align 4
-; CHECK-NEXT: store float [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 4
-; CHECK-NEXT: br label [[COMMON_RET]]
+; CHECK-NEXT: [[CALL1:%.*]] = tail call contract float @_Z3cosf(float [[X]])
+; CHECK-NEXT: store float [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT: ret void
;
entry:
br i1 %cond, label %bb0, label %bb1
@@ -607,18 +586,17 @@ bb1:
; The sin and cos are in different blocks but always execute
define void @sincos_f32_different_blocks_dominating_always_execute(i1 %cond, float %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out, ptr addrspace(1) %other) {
; CHECK-LABEL: define void @sincos_f32_different_blocks_dominating_always_execute
-; CHECK-SAME: (i1 [[COND:%.*]], float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[OTHER:%.*]]) local_unnamed_addr #[[ATTR3]] {
+; CHECK-SAME: (i1 [[COND:%.*]], float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[OTHER:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca float, align 4, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract float @_Z6sincosfPU3AS5f(float [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(5) [[__SINCOS_]], align 4
-; CHECK-NEXT: store float [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 4
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract float @_Z3sinf(float [[X]])
+; CHECK-NEXT: store float [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 4
; CHECK-NEXT: br i1 [[COND]], label [[BB0:%.*]], label [[BB1:%.*]]
; CHECK: bb0:
; CHECK-NEXT: store i32 0, ptr addrspace(1) [[OTHER]], align 4
; CHECK-NEXT: br label [[BB1]]
; CHECK: bb1:
-; CHECK-NEXT: store float [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT: [[CALL1:%.*]] = tail call contract float @_Z3cosf(float [[X]])
+; CHECK-NEXT: store float [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 4
; CHECK-NEXT: ret void
;
entry:
@@ -639,14 +617,17 @@ bb1:
; sin dominates cos but cos doesn't always execute.
define void @sincos_f32_different_blocks_dominating_conditional_execute(i1 %cond, float %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out, ptr addrspace(1) %other) {
; CHECK-LABEL: define void @sincos_f32_different_blocks_dominating_conditional_execute
-; CHECK-SAME: (i1 [[COND:%.*]], float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]], ptr addrspace(1) nocapture readnone [[OTHER:%.*]]) local_unnamed_addr #[[ATTR3]] {
+; CHECK-SAME: (i1 [[COND:%.*]], float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]], ptr addrspace(1) nocapture readnone [[OTHER:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca float, align 4, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract float @_Z6sincosfPU3AS5f(float [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(5) [[__SINCOS_]], align 4
-; CHECK-NEXT: store float [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 4
-; CHECK-NEXT: [[SPEC_SELECT:%.*]] = select i1 [[COND]], float [[TMP1]], float 0.000000e+00
-; CHECK-NEXT: store float [[SPEC_SELECT]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract float @_Z3sinf(float [[X]])
+; CHECK-NEXT: store float [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 4
+; CHECK-NEXT: br i1 [[COND]], label [[BB0:%.*]], label [[BB1:%.*]]
+; CHECK: bb0:
+; CHECK-NEXT: [[CALL1:%.*]] = tail call contract float @_Z3cosf(float [[X]])
+; CHECK-NEXT: br label [[BB1]]
+; CHECK: bb1:
+; CHECK-NEXT: [[PHI:%.*]] = phi float [ [[CALL1]], [[BB0]] ], [ 0.000000e+00, [[ENTRY:%.*]] ]
+; CHECK-NEXT: store float [[PHI]], ptr addrspace(1) [[COS_OUT]], align 4
; CHECK-NEXT: ret void
;
entry:
@@ -666,12 +647,11 @@ bb1:
define float @select_sin_or_cos_f32(i1 %cond, float %x) {
; CHECK-LABEL: define float @select_sin_or_cos_f32
-; CHECK-SAME: (i1 [[COND:%.*]], float [[X:%.*]]) local_unnamed_addr #[[ATTR5:[0-9]+]] {
+; CHECK-SAME: (i1 [[COND:%.*]], float [[X:%.*]]) #[[ATTR3:[0-9]+]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca float, align 4, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract float @_Z6sincosfPU3AS5f(float [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(5) [[__SINCOS_]], align 4
-; CHECK-NEXT: [[SELECT:%.*]] = select i1 [[COND]], float [[TMP0]], float [[TMP1]]
+; CHECK-NEXT: [[SIN:%.*]] = tail call contract float @_Z3sinf(float [[X]])
+; CHECK-NEXT: [[COS:%.*]] = tail call contract float @_Z3cosf(float [[X]])
+; CHECK-NEXT: [[SELECT:%.*]] = select i1 [[COND]], float [[SIN]], float [[COS]]
; CHECK-NEXT: ret float [[SELECT]]
;
entry:
@@ -685,15 +665,14 @@ declare void @func(ptr addrspace(1))
define void @sincos_f32_value_is_instr(ptr addrspace(1) %value.ptr, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_f32_value_is_instr
-; CHECK-SAME: (ptr addrspace(1) [[VALUE_PTR:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR4]] {
+; CHECK-SAME: (ptr addrspace(1) [[VALUE_PTR:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR4:[0-9]+]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca float, align 4, addrspace(5)
-; CHECK-NEXT: tail call void @func(ptr addrspace(1) [[VALUE_PTR]])
+; CHECK-NEXT: call void @func(ptr addrspace(1) [[VALUE_PTR]])
; CHECK-NEXT: [[X:%.*]] = load float, ptr addrspace(1) [[VALUE_PTR]], align 4
-; CHECK-NEXT: [[TMP0:%.*]] = call contract float @_Z6sincosfPU3AS5f(float [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(5) [[__SINCOS_]], align 4
-; CHECK-NEXT: store float [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 4
-; CHECK-NEXT: store float [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract float @_Z3sinf(float [[X]])
+; CHECK-NEXT: store float [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 4
+; CHECK-NEXT: [[CALL1:%.*]] = tail call contract float @_Z3cosf(float [[X]])
+; CHECK-NEXT: store float [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 4
; CHECK-NEXT: ret void
;
entry:
@@ -708,13 +687,12 @@ entry:
define void @sincos_f32_value_is_same_constexpr(ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_f32_value_is_same_constexpr
-; CHECK-SAME: (ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR3]] {
+; CHECK-SAME: (ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca float, align 4, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract float @_Z6sincosfPU3AS5f(float bitcast (i32 ptrtoint (ptr @func to i32) to float), ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(5) [[__SINCOS_]], align 4
-; CHECK-NEXT: store float [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 4
-; CHECK-NEXT: store float [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract float @_Z3sinf(float bitcast (i32 ptrtoint (ptr @func to i32) to float))
+; CHECK-NEXT: store float [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 4
+; CHECK-NEXT: [[CALL1:%.*]] = tail call contract float @_Z3cosf(float bitcast (i32 ptrtoint (ptr @func to i32) to float))
+; CHECK-NEXT: store float [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 4
; CHECK-NEXT: ret void
;
entry:
@@ -745,13 +723,12 @@ entry:
define void @sincos_f32_value_is_same_constantfp(ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_f32_value_is_same_constantfp
-; CHECK-SAME: (ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR3]] {
+; CHECK-SAME: (ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca float, align 4, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract float @_Z6sincosfPU3AS5f(float 4.200000e+01, ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(5) [[__SINCOS_]], align 4
-; CHECK-NEXT: store float [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 4
-; CHECK-NEXT: store float [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract float @_Z3sinf(float 4.200000e+01)
+; CHECK-NEXT: store float [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 4
+; CHECK-NEXT: [[CALL1:%.*]] = tail call contract float @_Z3cosf(float 4.200000e+01)
+; CHECK-NEXT: store float [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 4
; CHECK-NEXT: ret void
;
entry:
@@ -764,7 +741,7 @@ entry:
define void @sincos_f32_value_is_different_constantfp(ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_f32_value_is_different_constantfp
-; CHECK-SAME: (ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR2]] {
+; CHECK-SAME: (ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CALL:%.*]] = tail call contract float @_Z3sinf(float 4.200000e+01)
; CHECK-NEXT: store float [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 4
@@ -782,7 +759,7 @@ entry:
define void @sincos_f32_different_args(float %x, float %y, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_f32_different_args
-; CHECK-SAME: (float [[X:%.*]], float [[Y:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR2]] {
+; CHECK-SAME: (float [[X:%.*]], float [[Y:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CALL:%.*]] = tail call contract float @_Z3sinf(float [[X]])
; CHECK-NEXT: store float [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 4
@@ -800,13 +777,12 @@ entry:
define void @sincos_f32_flag_intersect0(float %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_f32_flag_intersect0
-; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR3]] {
+; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca float, align 4, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract float @_Z6sincosfPU3AS5f(float [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(5) [[__SINCOS_]], align 4
-; CHECK-NEXT: store float [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 4
-; CHECK-NEXT: store float [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT: [[CALL:%.*]] = tail call nnan contract float @_Z3sinf(float [[X]])
+; CHECK-NEXT: store float [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 4
+; CHECK-NEXT: [[CALL1:%.*]] = tail call ninf contract float @_Z3cosf(float [[X]])
+; CHECK-NEXT: store float [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 4
; CHECK-NEXT: ret void
;
entry:
@@ -819,13 +795,12 @@ entry:
define void @sincos_f32_flag_intersect1(float %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_f32_flag_intersect1
-; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR3]] {
+; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca float, align 4, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call nnan contract float @_Z6sincosfPU3AS5f(float [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(5) [[__SINCOS_]], align 4
-; CHECK-NEXT: store float [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 4
-; CHECK-NEXT: store float [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT: [[CALL:%.*]] = tail call nnan nsz contract float @_Z3sinf(float [[X]])
+; CHECK-NEXT: store float [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 4
+; CHECK-NEXT: [[CALL1:%.*]] = tail call nnan ninf contract float @_Z3cosf(float [[X]])
+; CHECK-NEXT: store float [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 4
; CHECK-NEXT: ret void
;
entry:
@@ -838,13 +813,12 @@ entry:
define void @sincos_v2f32_flag_intersect1(<2 x float> %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_v2f32_flag_intersect1
-; CHECK-SAME: (<2 x float> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR4]] {
+; CHECK-SAME: (<2 x float> [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca <2 x float>, align 8, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call nnan contract <2 x float> @_Z6sincosDv2_fPU3AS5S_(<2 x float> [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load <2 x float>, ptr addrspace(5) [[__SINCOS_]], align 8
-; CHECK-NEXT: store <2 x float> [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 8
-; CHECK-NEXT: store <2 x float> [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 8
+; CHECK-NEXT: [[CALL:%.*]] = tail call nnan ninf contract <2 x float> @_Z3sinDv2_f(<2 x float> [[X]])
+; CHECK-NEXT: store <2 x float> [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 8
+; CHECK-NEXT: [[CALL1:%.*]] = tail call nnan nsz contract <2 x float> @_Z3cosDv2_f(<2 x float> [[X]])
+; CHECK-NEXT: store <2 x float> [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 8
; CHECK-NEXT: ret void
;
entry:
@@ -859,15 +833,14 @@ declare void @use_stack_ptrs(ptr addrspace(5), ptr addrspace(5))
define void @sincos_f32_alloca_insertpt(float %x) {
; CHECK-LABEL: define void @sincos_f32_alloca_insertpt
-; CHECK-SAME: (float [[X:%.*]]) local_unnamed_addr #[[ATTR4]] {
+; CHECK-SAME: (float [[X:%.*]]) #[[ATTR4]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[ALLOCA0:%.*]] = alloca i32, align 4, addrspace(5)
; CHECK-NEXT: [[ALLOCA1:%.*]] = alloca i32, align 4, addrspace(5)
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca float, align 4, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract float @_Z6sincosfPU3AS5f(float [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(5) [[__SINCOS_]], align 4
-; CHECK-NEXT: store float [[TMP0]], ptr addrspace(5) [[ALLOCA0]], align 4
-; CHECK-NEXT: store float [[TMP1]], ptr addrspace(5) [[ALLOCA1]], align 4
+; CHECK-NEXT: [[SIN:%.*]] = tail call contract float @_Z3sinf(float [[X]])
+; CHECK-NEXT: [[COS:%.*]] = tail call contract float @_Z3cosf(float [[X]])
+; CHECK-NEXT: store float [[SIN]], ptr addrspace(5) [[ALLOCA0]], align 4
+; CHECK-NEXT: store float [[COS]], ptr addrspace(5) [[ALLOCA1]], align 4
; CHECK-NEXT: call void @use_stack_ptrs(ptr addrspace(5) [[ALLOCA0]], ptr addrspace(5) [[ALLOCA1]])
; CHECK-NEXT: ret void
;
@@ -884,9 +857,12 @@ entry:
define float @sincos_f32_unused_result_cos(float %x) {
; CHECK-LABEL: define float @sincos_f32_unused_result_cos
-; CHECK-SAME: (float [[X:%.*]]) local_unnamed_addr #[[ATTR6:[0-9]+]] {
+; CHECK-SAME: (float [[X:%.*]]) #[[ATTR3]] {
; CHECK-NEXT: entry:
+; CHECK-NEXT: [[ALLOCA0:%.*]] = alloca i32, align 4, addrspace(5)
+; CHECK-NEXT: [[ALLOCA1:%.*]] = alloca i32, align 4, addrspace(5)
; CHECK-NEXT: [[SIN:%.*]] = tail call contract float @_Z3sinf(float [[X]])
+; CHECK-NEXT: [[DEAD:%.*]] = tail call contract float @_Z3cosf(float [[X]])
; CHECK-NEXT: ret float [[SIN]]
;
entry:
@@ -899,8 +875,11 @@ entry:
define float @sincos_f32_unused_result_sin(float %x) {
; CHECK-LABEL: define float @sincos_f32_unused_result_sin
-; CHECK-SAME: (float [[X:%.*]]) local_unnamed_addr #[[ATTR6]] {
+; CHECK-SAME: (float [[X:%.*]]) #[[ATTR3]] {
; CHECK-NEXT: entry:
+; CHECK-NEXT: [[ALLOCA0:%.*]] = alloca i32, align 4, addrspace(5)
+; CHECK-NEXT: [[ALLOCA1:%.*]] = alloca i32, align 4, addrspace(5)
+; CHECK-NEXT: [[DEAD:%.*]] = tail call contract float @_Z3sinf(float [[X]])
; CHECK-NEXT: [[COS:%.*]] = tail call contract float @_Z3cosf(float [[X]])
; CHECK-NEXT: ret float [[COS]]
;
@@ -914,16 +893,18 @@ entry:
define void @sincos_f32_repeated_uses(float %x, ptr addrspace(1) %sin_out, ptr addrspace(1) %cos_out) {
; CHECK-LABEL: define void @sincos_f32_repeated_uses
-; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) [[SIN_OUT:%.*]], ptr addrspace(1) [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR7:[0-9]+]] {
+; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) [[SIN_OUT:%.*]], ptr addrspace(1) [[COS_OUT:%.*]]) #[[ATTR5:[0-9]+]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca float, align 4, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract float @_Z6sincosfPU3AS5f(float [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(5) [[__SINCOS_]], align 4
-; CHECK-NEXT: store volatile float [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 4
-; CHECK-NEXT: store volatile float [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 4
-; CHECK-NEXT: store volatile float [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 4
-; CHECK-NEXT: store volatile float [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 4
-; CHECK-NEXT: store volatile float [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 4
+; CHECK-NEXT: [[SIN0:%.*]] = tail call contract float @_Z3sinf(float [[X]])
+; CHECK-NEXT: store volatile float [[SIN0]], ptr addrspace(1) [[SIN_OUT]], align 4
+; CHECK-NEXT: [[SIN1:%.*]] = tail call contract float @_Z3sinf(float [[X]])
+; CHECK-NEXT: store volatile float [[SIN1]], ptr addrspace(1) [[SIN_OUT]], align 4
+; CHECK-NEXT: [[COS0:%.*]] = tail call contract float @_Z3cosf(float [[X]])
+; CHECK-NEXT: store volatile float [[COS0]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT: [[COS1:%.*]] = tail call contract float @_Z3cosf(float [[X]])
+; CHECK-NEXT: store volatile float [[COS1]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT: [[SIN2:%.*]] = tail call contract float @_Z3sinf(float [[X]])
+; CHECK-NEXT: store volatile float [[SIN2]], ptr addrspace(1) [[SIN_OUT]], align 4
; CHECK-NEXT: ret void
;
entry:
@@ -947,7 +928,7 @@ entry:
define void @sin_f32_indirect_call_user(float %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out, ptr %func.ptr) {
; CHECK-LABEL: define void @sin_f32_indirect_call_user
-; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]], ptr nocapture readonly [[FUNC_PTR:%.*]]) local_unnamed_addr #[[ATTR4]] {
+; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]], ptr nocapture readonly [[FUNC_PTR:%.*]]) #[[ATTR4]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CALL:%.*]] = tail call contract float @_Z3sinf(float [[X]])
; CHECK-NEXT: store float [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 4
@@ -965,7 +946,7 @@ entry:
define void @cos_f32_indirect_call_user(float %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out, ptr %func.ptr) {
; CHECK-LABEL: define void @cos_f32_indirect_call_user
-; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]], ptr nocapture readonly [[FUNC_PTR:%.*]]) local_unnamed_addr #[[ATTR4]] {
+; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]], ptr nocapture readonly [[FUNC_PTR:%.*]]) #[[ATTR4]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CALL:%.*]] = tail call contract float @_Z3cosf(float [[X]])
; CHECK-NEXT: store float [[CALL]], ptr addrspace(1) [[COS_OUT]], align 4
@@ -983,13 +964,12 @@ entry:
define void @sincos_f32_preserve_fpmath_0(float %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_f32_preserve_fpmath_0
-; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR3]] {
+; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca float, align 4, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract float @_Z6sincosfPU3AS5f(float [[X]], ptr addrspace(5) [[__SINCOS_]]), !fpmath [[META5:![0-9]+]]
-; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(5) [[__SINCOS_]], align 4
-; CHECK-NEXT: store float [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 4
-; CHECK-NEXT: store float [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract float @_Z3sinf(float [[X]]), !fpmath [[META5:![0-9]+]]
+; CHECK-NEXT: store float [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 4
+; CHECK-NEXT: [[CALL1:%.*]] = tail call contract float @_Z3cosf(float [[X]]), !fpmath [[META5]]
+; CHECK-NEXT: store float [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 4
; CHECK-NEXT: ret void
;
entry:
@@ -1002,13 +982,12 @@ entry:
define void @sincos_f32_preserve_fpmath_1(float %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_f32_preserve_fpmath_1
-; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR3]] {
+; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca float, align 4, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract float @_Z6sincosfPU3AS5f(float [[X]], ptr addrspace(5) [[__SINCOS_]]), !fpmath [[META6:![0-9]+]]
-; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(5) [[__SINCOS_]], align 4
-; CHECK-NEXT: store float [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 4
-; CHECK-NEXT: store float [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract float @_Z3sinf(float [[X]]), !fpmath [[META5]]
+; CHECK-NEXT: store float [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 4
+; CHECK-NEXT: [[CALL1:%.*]] = tail call contract float @_Z3cosf(float [[X]]), !fpmath [[META6:![0-9]+]]
+; CHECK-NEXT: store float [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 4
; CHECK-NEXT: ret void
;
entry:
@@ -1022,13 +1001,12 @@ entry:
; Should drop the metadata
define void @sincos_f32_drop_fpmath(float %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
; CHECK-LABEL: define void @sincos_f32_drop_fpmath
-; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR3]] {
+; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca float, align 4, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract float @_Z6sincosfPU3AS5f(float [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(5) [[__SINCOS_]], align 4
-; CHECK-NEXT: store float [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 4
-; CHECK-NEXT: store float [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract float @_Z3sinf(float [[X]]), !fpmath [[META5]]
+; CHECK-NEXT: store float [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 4
+; CHECK-NEXT: [[CALL1:%.*]] = tail call contract float @_Z3cosf(float [[X]])
+; CHECK-NEXT: store float [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 4
; CHECK-NEXT: ret void
;
entry:
@@ -1041,16 +1019,15 @@ entry:
define void @sincos_f32_debuginfo(float %x, ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) !dbg !15 {
; CHECK-LABEL: define void @sincos_f32_debuginfo
-; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR3]] !dbg [[DBG7:![0-9]+]] {
+; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR2]] !dbg [[DBG7:![0-9]+]] {
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca float, align 4, addrspace(5), !dbg [[DBG14:![0-9]+]]
-; CHECK-NEXT: [[TMP0:%.*]] = call contract float @_Z6sincosfPU3AS5f(float [[X]], ptr addrspace(5) [[__SINCOS_]]), !dbg [[DBG14]]
-; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(5) [[__SINCOS_]], align 4, !dbg [[DBG14]]
-; CHECK-NEXT: #dbg_value(float [[TMP0]], [[META11:![0-9]+]], !DIExpression(), [[META15:![0-9]+]])
-; CHECK-NEXT: store float [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 4, !dbg [[DBG16:![0-9]+]]
-; CHECK-NEXT: #dbg_value(float [[TMP1]], [[META13:![0-9]+]], !DIExpression(), [[META17:![0-9]+]])
-; CHECK-NEXT: store float [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 4, !dbg [[DBG18:![0-9]+]]
-; CHECK-NEXT: ret void, !dbg [[DBG19:![0-9]+]]
+; CHECK-NEXT: [[CALL:%.*]] = tail call contract float @_Z3sinf(float [[X]]), !dbg [[DBG14:![0-9]+]]
+; CHECK-NEXT: #dbg_value(float [[CALL]], [[META11:![0-9]+]], !DIExpression(), [[DBG14]])
+; CHECK-NEXT: store float [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 4, !dbg [[DBG15:![0-9]+]]
+; CHECK-NEXT: [[CALL1:%.*]] = tail call contract float @_Z3cosf(float [[X]]), !dbg [[DBG16:![0-9]+]]
+; CHECK-NEXT: #dbg_value(float [[CALL1]], [[META13:![0-9]+]], !DIExpression(), [[DBG16]])
+; CHECK-NEXT: store float [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 4, !dbg [[DBG17:![0-9]+]]
+; CHECK-NEXT: ret void, !dbg [[DBG18:![0-9]+]]
;
entry:
%call = tail call contract float @_Z3sinf(float %x), !dbg !19
@@ -1064,7 +1041,7 @@ entry:
define float @sin_sincos_private_f32(float %x, ptr addrspace(1) %sin_out, ptr addrspace(1) %cos_out) {
; CHECK-LABEL: define float @sin_sincos_private_f32
-; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR3]] {
+; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR6:[0-9]+]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[COS_TMP:%.*]] = alloca float, align 4, addrspace(5)
; CHECK-NEXT: [[SIN0:%.*]] = tail call nnan ninf nsz contract float @_Z3sinf(float [[X]]), !fpmath [[META5]]
@@ -1086,7 +1063,7 @@ entry:
define float @sin_sincos_generic_f32(float %x, ptr addrspace(1) %sin_out, ptr addrspace(1) %cos_out) {
; CHECK-LABEL: define float @sin_sincos_generic_f32
-; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR3]] {
+; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR6]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[COS_TMP:%.*]] = alloca float, align 4, addrspace(5)
; CHECK-NEXT: [[SIN0:%.*]] = tail call nsz contract float @_Z3sinf(float [[X]]), !fpmath [[META5]]
@@ -1110,7 +1087,7 @@ entry:
define float @cos_sincos_private_f32(float %x, ptr addrspace(1) %sin_out, ptr addrspace(1) %cos_out) {
; CHECK-LABEL: define float @cos_sincos_private_f32
-; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture readnone [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR3]] {
+; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture readnone [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR6]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[COS_TMP:%.*]] = alloca float, align 4, addrspace(5)
; CHECK-NEXT: [[COS0:%.*]] = tail call contract float @_Z3cosf(float [[X]])
@@ -1132,7 +1109,7 @@ entry:
define float @cos_sincos_generic_f32(float %x, ptr addrspace(1) %sin_out, ptr addrspace(1) %cos_out) {
; CHECK-LABEL: define float @cos_sincos_generic_f32
-; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture readnone [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR3]] {
+; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture readnone [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR6]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[COS_TMP:%.*]] = alloca float, align 4, addrspace(5)
; CHECK-NEXT: [[COS0:%.*]] = tail call contract float @_Z3cosf(float [[X]])
@@ -1156,7 +1133,7 @@ entry:
define float @sincos_private_f32_x2(float %x, ptr addrspace(1) %sin_out, ptr addrspace(1) %cos_out) {
; CHECK-LABEL: define float @sincos_private_f32_x2
-; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture readnone [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR3]] {
+; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture readnone [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR6]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[COS_TMP0:%.*]] = alloca float, align 4, addrspace(5)
; CHECK-NEXT: [[COS_TMP1:%.*]] = alloca float, align 4, addrspace(5)
@@ -1184,7 +1161,7 @@ entry:
define float @sincos_generic_f32_x2(float %x, ptr addrspace(1) %sin_out, ptr addrspace(1) %cos_out) {
; CHECK-LABEL: define float @sincos_generic_f32_x2
-; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture readnone [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR3]] {
+; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture readnone [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR6]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[COS_TMP0:%.*]] = alloca float, align 4, addrspace(5)
; CHECK-NEXT: [[COS_TMP1:%.*]] = alloca float, align 4, addrspace(5)
@@ -1194,7 +1171,8 @@ define float @sincos_generic_f32_x2(float %x, ptr addrspace(1) %sin_out, ptr add
; CHECK-NEXT: store float [[COS0]], ptr addrspace(1) [[COS_OUT]], align 4
; CHECK-NEXT: [[COS_TMP1_CAST:%.*]] = addrspacecast ptr addrspace(5) [[COS_TMP1]] to ptr
; CHECK-NEXT: [[SIN1:%.*]] = call contract float @_Z6sincosfPU3AS0f(float [[X]], ptr [[COS_TMP1_CAST]])
-; CHECK-NEXT: store float [[COS0]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT: [[COS1:%.*]] = load float, ptr addrspace(5) [[COS_TMP0]], align 4
+; CHECK-NEXT: store float [[COS1]], ptr addrspace(1) [[COS_OUT]], align 4
; CHECK-NEXT: ret float [[SIN1]]
;
entry:
@@ -1213,7 +1191,7 @@ entry:
define float @sincos_generic_private_f32(float %x, ptr addrspace(1) %sin_out, ptr addrspace(1) %cos_out) {
; CHECK-LABEL: define float @sincos_generic_private_f32
-; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture readnone [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR3]] {
+; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture readnone [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR6]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[COS_TMP0:%.*]] = alloca float, align 4, addrspace(5)
; CHECK-NEXT: [[COS_TMP1:%.*]] = alloca float, align 4, addrspace(5)
@@ -1222,7 +1200,8 @@ define float @sincos_generic_private_f32(float %x, ptr addrspace(1) %sin_out, pt
; CHECK-NEXT: store float [[COS0]], ptr addrspace(1) [[COS_OUT]], align 4
; CHECK-NEXT: [[COS_TMP1_CAST:%.*]] = addrspacecast ptr addrspace(5) [[COS_TMP1]] to ptr
; CHECK-NEXT: [[SIN1:%.*]] = call contract float @_Z6sincosfPU3AS0f(float [[X]], ptr [[COS_TMP1_CAST]])
-; CHECK-NEXT: store float [[COS0]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT: [[COS1:%.*]] = load float, ptr addrspace(5) [[COS_TMP0]], align 4
+; CHECK-NEXT: store float [[COS1]], ptr addrspace(1) [[COS_OUT]], align 4
; CHECK-NEXT: ret float [[SIN1]]
;
entry:
@@ -1240,18 +1219,22 @@ entry:
define float @sincos_mixed_sin_cos_generic_private_f32(float %x, ptr addrspace(1) %sin_out, ptr addrspace(1) %cos_out) {
; CHECK-LABEL: define float @sincos_mixed_sin_cos_generic_private_f32
-; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR3]] {
+; CHECK-SAME: (float [[X:%.*]], ptr addrspace(1) nocapture writeonly [[SIN_OUT:%.*]], ptr addrspace(1) nocapture writeonly [[COS_OUT:%.*]]) #[[ATTR6]] {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[COS_TMP0:%.*]] = alloca float, align 4, addrspace(5)
-; CHECK-NEXT: [[__SINCOS_:%.*]] = alloca float, align 4, addrspace(5)
-; CHECK-NEXT: [[TMP0:%.*]] = call contract float @_Z6sincosfPU3AS5f(float [[X]], ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr addrspace(5) [[__SINCOS_]], align 4
+; CHECK-NEXT: [[COS_TMP1:%.*]] = alloca float, align 4, addrspace(5)
; CHECK-NEXT: [[SIN0:%.*]] = call contract float @_Z6sincosfPU3AS5f(float [[X]], ptr addrspace(5) [[COS_TMP0]])
-; CHECK-NEXT: store float [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 4
-; CHECK-NEXT: store float [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT: [[SIN1:%.*]] = call contract float @_Z3sinf(float [[X]])
+; CHECK-NEXT: store float [[SIN1]], ptr addrspace(1) [[SIN_OUT]], align 4
+; CHECK-NEXT: [[COS0:%.*]] = load float, ptr addrspace(5) [[COS_TMP0]], align 4
+; CHECK-NEXT: store float [[COS0]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT: [[COS_TMP1_CAST:%.*]] = addrspacecast ptr addrspace(5) [[COS_TMP1]] to ptr
+; CHECK-NEXT: [[COS1:%.*]] = call contract float @_Z3cosf(float [[X]])
+; CHECK-NEXT: store float [[COS1]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT: [[SIN2:%.*]] = call contract float @_Z6sincosfPU3AS0f(float [[X]], ptr [[COS_TMP1_CAST]])
; CHECK-NEXT: [[COS2:%.*]] = load float, ptr addrspace(5) [[COS_TMP0]], align 4
; CHECK-NEXT: store float [[COS2]], ptr addrspace(1) [[COS_OUT]], align 4
-; CHECK-NEXT: ret float [[TMP0]]
+; CHECK-NEXT: ret float [[SIN2]]
;
entry:
%cos.tmp0 = alloca float, addrspace(5)
diff --git a/llvm/test/CodeGen/AMDGPU/attributor-noopt.ll b/llvm/test/CodeGen/AMDGPU/attributor-noopt.ll
index 90562e25a3e9c..9ce0f55255b8f 100644
--- a/llvm/test/CodeGen/AMDGPU/attributor-noopt.ll
+++ b/llvm/test/CodeGen/AMDGPU/attributor-noopt.ll
@@ -1,4 +1,4 @@
-; RUN: sed 's/CODE_OBJECT_VERSION/500/g' %s | opt -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -O2 | llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 | FileCheck -check-prefix=OPT %s
+; RUN: sed 's/CODE_OBJECT_VERSION/500/g' %s | opt -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -passes="lto<O2>" | llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 | FileCheck -check-prefix=OPT %s
; RUN: sed 's/CODE_OBJECT_VERSION/400/g' %s | opt -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -O0 | llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 | FileCheck -check-prefixes=NOOPT,COV4 %s
; RUN: sed 's/CODE_OBJECT_VERSION/500/g' %s | opt -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -O0 | llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 | FileCheck -check-prefixes=NOOPT,COV5 %s
diff --git a/llvm/test/CodeGen/AMDGPU/inline-attr.ll b/llvm/test/CodeGen/AMDGPU/inline-attr.ll
index e7a7b8a335d0d..55314fc64f0f1 100644
--- a/llvm/test/CodeGen/AMDGPU/inline-attr.ll
+++ b/llvm/test/CodeGen/AMDGPU/inline-attr.ll
@@ -1,6 +1,6 @@
-; RUN: opt -mtriple=amdgcn--amdhsa -S -O3 -enable-unsafe-fp-math %s | FileCheck -check-prefix=GCN -check-prefix=UNSAFE %s
-; RUN: opt -mtriple=amdgcn--amdhsa -S -O3 -enable-no-nans-fp-math %s | FileCheck -check-prefix=GCN -check-prefix=NONANS %s
-; RUN: opt -mtriple=amdgcn--amdhsa -S -O3 -enable-no-infs-fp-math %s | FileCheck -check-prefix=GCN -check-prefix=NOINFS %s
+; RUN: opt -mtriple=amdgcn--amdhsa -S -passes="lto<O3>" -enable-unsafe-fp-math %s | FileCheck -check-prefix=GCN -check-prefix=UNSAFE %s
+; RUN: opt -mtriple=amdgcn--amdhsa -S -passes="lto<O3>" -enable-no-nans-fp-math %s | FileCheck -check-prefix=GCN -check-prefix=NONANS %s
+; RUN: opt -mtriple=amdgcn--amdhsa -S -passes="lto<O3>" -enable-no-infs-fp-math %s | FileCheck -check-prefix=GCN -check-prefix=NOINFS %s
; GCN: define float @foo(float %x) local_unnamed_addr #0 {
; GCN: define amdgpu_kernel void @caller(ptr addrspace(1) nocapture %p) local_unnamed_addr #1 {
diff --git a/llvm/test/CodeGen/AMDGPU/simplify-libcalls.ll b/llvm/test/CodeGen/AMDGPU/simplify-libcalls.ll
index bf21ed6898ff4..ecc03feff2bf4 100644
--- a/llvm/test/CodeGen/AMDGPU/simplify-libcalls.ll
+++ b/llvm/test/CodeGen/AMDGPU/simplify-libcalls.ll
@@ -826,5 +826,5 @@ entry:
; GCN-PRELINK: declare float @_Z4cbrtf(float) local_unnamed_addr #[[$NOUNWIND_READONLY:[0-9]+]]
; GCN-PRELINK-DAG: attributes #[[$NOUNWIND]] = { nounwind }
-; GCN-PRELINK-DAG: attributes #[[$NOUNWIND_READONLY]] = { nounwind memory(read) "amdgpu-waves-per-eu"="4,10" "uniform-work-group-size"="false" }
+; GCN-PRELINK-DAG: attributes #[[$NOUNWIND_READONLY]] = { nounwind memory(read) }
attributes #0 = { nounwind }
More information about the llvm-commits
mailing list