0dd21ad1c6
This just added unnecessary work to the IR, since they are only used for load and store, which just causes some IR noise. Tests updated by UTC script to remove the extra lines.
134 lines
6.0 KiB
C++
134 lines
6.0 KiB
C++
// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --check-globals all --version 6
|
|
// REQUIRES: amdgpu-registered-target
|
|
// RUN: %clang_cc1 -x hip -triple x86_64-unknown-linux-gnu -aux-triple amdgcn-amd-amdhsa -emit-llvm-bc %s -o %t-host.bc
|
|
// RUN: %clang_cc1 -x hip -triple x86_64-unknown-linux-gnu -aux-triple amdgcn-amd-amdhsa -emit-llvm %s -o - | FileCheck %s --check-prefix=HOST
|
|
// RUN: %clang_cc1 -x hip -triple amdgcn-amd-amdhsa -aux-triple x86_64-unknown-linux-gnu -emit-llvm %s -fcuda-is-device -o - | FileCheck %s --check-prefix=DEVICE
|
|
|
|
#define __device__ __attribute__((device))
|
|
#define __host__ __attribute__((host))
|
|
|
|
extern "C" {
|
|
//.
|
|
// HOST: @__HostVar = global i32 1, align 4
|
|
// HOST: @__hip_cuid_ = global i8 0
|
|
// HOST: @llvm.compiler.used = appending global [1 x ptr] [ptr @__hip_cuid_], section "llvm.metadata"
|
|
// HOST: @HostFunc = weak alias i32 (), ptr @__HostFunc
|
|
// HOST: @HostFunc_ = alias i32 (), ptr @__HostFunc
|
|
// HOST: @HostVar = weak alias i32, ptr @__HostVar
|
|
// HOST: @HostVar_ = alias i32, ptr @__HostVar
|
|
// HOST: @Two = weak alias i32 (), ptr @__Two
|
|
// HOST: @Two_ = alias i32 (), ptr @__Two
|
|
// HOST: @_Z5Threev = weak alias i32 (), ptr @_Z7__Threev
|
|
// HOST: @_Z6Three_v = alias i32 (), ptr @_Z7__Threev
|
|
// HOST: @_Z4Fourv = weak alias i32 (), ptr @_Z6__Fourv
|
|
// HOST: @_Z4Fourf = weak alias float (float), ptr @_Z6__Fourf
|
|
//.
|
|
// DEVICE: @__hip_cuid_ = addrspace(1) global i8 0
|
|
// DEVICE: @llvm.compiler.used = appending addrspace(1) global [1 x ptr] [ptr addrspacecast (ptr addrspace(1) @__hip_cuid_ to ptr)], section "llvm.metadata"
|
|
// DEVICE: @One = weak alias i32 (), ptr @__One
|
|
// DEVICE: @One_ = alias i32 (), ptr @__One
|
|
// DEVICE: @Two = weak alias i32 (), ptr @__Two
|
|
// DEVICE: @Two_ = alias i32 (), ptr @__Two
|
|
// DEVICE: @_Z5Threev = weak alias i32 (), ptr @_Z7__Threev
|
|
// DEVICE: @_Z6Three_v = alias i32 (), ptr @_Z7__Threev
|
|
// DEVICE: @_Z4Fourv = weak alias i32 (), ptr @_Z6__Fourv
|
|
// DEVICE: @_Z4Fourf = weak alias float (float), ptr @_Z6__Fourf
|
|
//.
|
|
// HOST-LABEL: define dso_local i32 @__HostFunc(
|
|
// HOST-SAME: ) #[[ATTR0:[0-9]+]] {
|
|
// HOST-NEXT: [[ENTRY:.*:]]
|
|
// HOST-NEXT: ret i32 42
|
|
//
|
|
int __HostFunc(void) { return 42; }
|
|
int __HostVar = 1;
|
|
int HostFunc(void) __attribute__((weak, alias("__HostFunc")));
|
|
int HostFunc_(void) __attribute__((alias("__HostFunc")));
|
|
extern int __attribute__((weak, alias("__HostVar"))) HostVar;
|
|
extern int __attribute__((alias("__HostVar"))) HostVar_;
|
|
|
|
// DEVICE-LABEL: define dso_local i32 @__One(
|
|
// DEVICE-SAME: ) #[[ATTR0:[0-9]+]] {
|
|
// DEVICE-NEXT: [[ENTRY:.*:]]
|
|
// DEVICE-NEXT: ret i32 1
|
|
//
|
|
__device__ int __One(void) { return 1; }
|
|
__device__ int One(void) __attribute__((weak, alias("__One")));
|
|
__device__ int One_(void) __attribute__((alias("__One")));
|
|
|
|
// HOST-LABEL: define dso_local i32 @__Two(
|
|
// HOST-SAME: ) #[[ATTR0]] {
|
|
// HOST-NEXT: [[ENTRY:.*:]]
|
|
// HOST-NEXT: ret i32 2
|
|
//
|
|
// DEVICE-LABEL: define dso_local i32 @__Two(
|
|
// DEVICE-SAME: ) #[[ATTR0]] {
|
|
// DEVICE-NEXT: [[ENTRY:.*:]]
|
|
// DEVICE-NEXT: ret i32 2
|
|
//
|
|
__host__ __device__ int __Two(void) { return 2; }
|
|
__host__ __device__ int Two(void) __attribute__((weak, alias("__Two")));
|
|
__host__ __device__ int Two_(void) __attribute__((alias("__Two")));
|
|
}
|
|
|
|
// HOST-LABEL: define linkonce_odr noundef i32 @_Z7__Threev(
|
|
// HOST-SAME: ) #[[ATTR0]] comdat {
|
|
// HOST-NEXT: [[ENTRY:.*:]]
|
|
// HOST-NEXT: ret i32 5
|
|
//
|
|
// DEVICE-LABEL: define linkonce_odr noundef i32 @_Z7__Threev(
|
|
// DEVICE-SAME: ) #[[ATTR0]] comdat {
|
|
// DEVICE-NEXT: [[ENTRY:.*:]]
|
|
// DEVICE-NEXT: ret i32 5
|
|
//
|
|
__host__ __device__ constexpr int __Three(void) { return 5; }
|
|
__host__ __device__ int Three(void) __attribute__((weak, alias("_Z7__Threev")));
|
|
__host__ __device__ int Three_(void) __attribute__((alias("_Z7__Threev")));
|
|
|
|
|
|
// HOST-LABEL: define dso_local noundef i32 @_Z6__Fourv(
|
|
// HOST-SAME: ) #[[ATTR0]] {
|
|
// HOST-NEXT: [[ENTRY:.*:]]
|
|
// HOST-NEXT: ret i32 2
|
|
//
|
|
// DEVICE-LABEL: define dso_local noundef i32 @_Z6__Fourv(
|
|
// DEVICE-SAME: ) #[[ATTR0]] {
|
|
// DEVICE-NEXT: [[ENTRY:.*:]]
|
|
// DEVICE-NEXT: ret i32 2
|
|
//
|
|
__host__ __device__ int __Four(void) { return 2; }
|
|
// HOST-LABEL: define dso_local noundef float @_Z6__Fourf(
|
|
// HOST-SAME: float noundef [[F:%.*]]) #[[ATTR0]] {
|
|
// HOST-NEXT: [[ENTRY:.*:]]
|
|
// HOST-NEXT: [[F_ADDR:%.*]] = alloca float, align 4
|
|
// HOST-NEXT: store float [[F]], ptr [[F_ADDR]], align 4
|
|
// HOST-NEXT: [[TMP0:%.*]] = load float, ptr [[F_ADDR]], align 4
|
|
// HOST-NEXT: [[MUL:%.*]] = fmul contract float 2.000000e+00, [[TMP0]]
|
|
// HOST-NEXT: ret float [[MUL]]
|
|
//
|
|
// DEVICE-LABEL: define dso_local noundef float @_Z6__Fourf(
|
|
// DEVICE-SAME: float noundef [[F:%.*]]) #[[ATTR0]] {
|
|
// DEVICE-NEXT: [[ENTRY:.*:]]
|
|
// DEVICE-NEXT: [[F_ADDR:%.*]] = alloca float, align 4, addrspace(5)
|
|
// DEVICE-NEXT: [[F_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[F_ADDR]] to ptr
|
|
// DEVICE-NEXT: store float [[F]], ptr [[F_ADDR_ASCAST]], align 4
|
|
// DEVICE-NEXT: [[TMP0:%.*]] = load float, ptr [[F_ADDR_ASCAST]], align 4
|
|
// DEVICE-NEXT: [[MUL:%.*]] = fmul contract float 2.000000e+00, [[TMP0]]
|
|
// DEVICE-NEXT: ret float [[MUL]]
|
|
//
|
|
__host__ __device__ float __Four(float f) { return 2.0f * f; }
|
|
__host__ __device__ int Four(void) __attribute__((weak, alias("_Z6__Fourv")));
|
|
__host__ __device__ float Four(float f) __attribute__((weak, alias("_Z6__Fourf")));
|
|
//.
|
|
// HOST: attributes #[[ATTR0]] = { mustprogress noinline nounwind optnone "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+cx8,+mmx,+sse,+sse2,+x87" }
|
|
//.
|
|
// DEVICE: attributes #[[ATTR0]] = { convergent mustprogress noinline nounwind optnone "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
|
|
//.
|
|
// HOST: [[META0:![0-9]+]] = !{i32 1, !"wchar_size", i32 4}
|
|
// HOST: [[META1:![0-9]+]] = !{!"{{.*}}clang version {{.*}}"}
|
|
//.
|
|
// DEVICE: [[META0:![0-9]+]] = !{i32 1, !"amdhsa_code_object_version", i32 600}
|
|
// DEVICE: [[META1:![0-9]+]] = !{i32 1, !"amdgpu_printf_kind", !"hostcall"}
|
|
// DEVICE: [[META2:![0-9]+]] = !{i32 1, !"wchar_size", i32 4}
|
|
// DEVICE: [[META3:![0-9]+]] = !{!"{{.*}}clang version {{.*}}"}
|
|
//.
|