clang: Add start of header test for __clang_hip_libdevice_declares

[arsenm]

Summary

It was shockingly difficult to get identical checks out of the 3
languages. The visibility, dso_local and fp-contract settings seem to
differ unless I force every test function to
static. update_cc_test_checks also isn't smart enough when the
attribute groups slightly differ.

[arsenm]

Use -verify

[JonChesterfield]

That seems like something we should fix. Visibility/dso really should be the same default. I don't know what that fp thing is

[arsenm]

In D152829#4419016, @JonChesterfield wrote:

That seems like something we should fix. Visibility/dso really should be the same default. I don't know what that fp thing is

Whether you're allowed to turn x * y + z into an fma if you feel like it.

[arsenm]

ping

[arsenm]

40bb302c451ec1a8f6a2b8238e0a56448b8e1a12

clang/test/Headers/__clang_hip_libdevice_declares.cpp

+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 2
+// REQUIRES: amdgpu-registered-target, x86-registered-target
+
+// fp-contract, -no-enable-noundef-analysis and visibility are to just get the
+// same output for openmp and hip.
+
+// RUN: %clang_cc1 -x c++ -fopenmp -fvisibility=default -ffp-contract=off -no-enable-noundef-analysis \
+// RUN:   -internal-isystem %S/../../lib/Headers/openmp_wrappers  \
+// RUN:   -include __clang_openmp_device_functions.h \
+// RUN:   -internal-isystem %S/../../lib/Headers/openmp_wrappers \
+// RUN:   -internal-isystem %S/Inputs/include \
+// RUN:   -triple amdgcn-amd-amdhsa -aux-triple x86_64-unknown-unknown -verify \
+// RUN:   -fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm -fopenmp-is-device \
+// RUN:   -o - %s | FileCheck --check-prefixes=CHECK,OPENMP,OPENMP-CPP %s
+
+// RUN: %clang_cc1 -x c -fopenmp -fvisibility=default -ffp-contract=off -no-enable-noundef-analysis \
+// RUN:   -internal-isystem %S/../../lib/Headers/openmp_wrappers  \
+// RUN:   -include __clang_openmp_device_functions.h \
+// RUN:   -internal-isystem %S/../../lib/Headers/openmp_wrappers \
+// RUN:   -internal-isystem %S/Inputs/include \
+// RUN:   -triple amdgcn-amd-amdhsa -aux-triple x86_64-unknown-unknown -verify \
+// RUN:   -fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm -fopenmp-is-device \
+// RUN:   -o - %s | FileCheck --check-prefixes=CHECK,OPENMP,OPENMP-C %s
+
+// RUN: %clang_cc1 -x hip -fvisibility=default -ffp-contract=off -no-enable-noundef-analysis \
+// RUN:   -include __clang_hip_runtime_wrapper.h  \
+// RUN:   -internal-isystem %S/../../lib/Headers/cuda_wrappers \
+// RUN:   -internal-isystem %S/Inputs/include \
+// RUN:   -triple amdgcn-amd-amdhsa -aux-triple x86_64-unknown-unknown -verify \
+// RUN:   -emit-llvm -fcuda-is-device -o - \
+// RUN:   -D__HIPCC_RTC__ %s | FileCheck -check-prefixes=CHECK,HIP %s
+
+// expected-no-diagnostics
+
+#ifdef __cplusplus
+#include <cmath>
+#else
+#include <math.h>
+#endif
+
+#ifdef _OPENMP
+#define __device__
+#endif
+
+// static and overloadable to get the same function annotations between C, C++ and HIP.
+#define TEST_FUNC_ATTRS static __device__ __attribute__((used,overloadable))
+
+#ifdef _OPENMP
+#pragma omp begin declare target
+#endif
+
+// This function is a hack to get the same IR out of HIP and OpenMP. The HIP
+// headers declare __cxa_* functions with these attributes, such that the
+// attribute groups are different. update_cc_test_checks isn't smart enough to
+// strip attributes from the checks, or semantically compare them.
+__attribute__((__visibility__("default")))
+__attribute__((weak))
+__attribute__((noreturn))
+__attribute__((overloadable))
+// OPENMP-LABEL: define weak hidden void @_Z20attribute_check_hackv
+// OPENMP-SAME: () #[[ATTR0:[0-9]+]] {
+// OPENMP-NEXT:  entry:
+// OPENMP-NEXT:    call void @llvm.trap()
+// OPENMP-NEXT:    unreachable
+//
+void attribute_check_hack(void) {
+  __builtin_trap();
+}
+
+// CHECK-LABEL: define internal float @_ZL18test_ockl_acos_f32f
+// CHECK-SAME: (float [[SRC:%.*]]) #[[ATTR2:[0-9]+]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca float, align 4, addrspace(5)
+// CHECK-NEXT:    [[SRC_ADDR:%.*]] = alloca float, align 4, addrspace(5)
+// CHECK-NEXT:    [[RETVAL_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL]] to ptr
+// CHECK-NEXT:    [[SRC_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[SRC_ADDR]] to ptr
+// CHECK-NEXT:    store float [[SRC]], ptr [[SRC_ADDR_ASCAST]], align 4
+// CHECK-NEXT:    [[TMP0:%.*]] = load float, ptr [[SRC_ADDR_ASCAST]], align 4
+// CHECK-NEXT:    [[CALL:%.*]] = call float @__ocml_acos_f32(float [[TMP0]]) #[[ATTR4:[0-9]+]]
+// CHECK-NEXT:    ret float [[CALL]]
+//
+TEST_FUNC_ATTRS float test_ockl_acos_f32(float src) {
+  return __ocml_acos_f32(src);
+}
+
+// CHECK-LABEL: define internal float @_ZL15test_ockl_fdot2Dv2_DF16_S_fbi
+// CHECK-SAME: (<2 x half> [[A:%.*]], <2 x half> [[B:%.*]], float [[C:%.*]], i1 zeroext [[S:%.*]], i32 [[S_INT:%.*]]) #[[ATTR2]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca float, align 4, addrspace(5)
+// CHECK-NEXT:    [[A_ADDR:%.*]] = alloca <2 x half>, align 4, addrspace(5)
+// CHECK-NEXT:    [[B_ADDR:%.*]] = alloca <2 x half>, align 4, addrspace(5)
+// CHECK-NEXT:    [[C_ADDR:%.*]] = alloca float, align 4, addrspace(5)
+// CHECK-NEXT:    [[S_ADDR:%.*]] = alloca i8, align 1, addrspace(5)
+// CHECK-NEXT:    [[S_INT_ADDR:%.*]] = alloca i32, align 4, addrspace(5)
+// CHECK-NEXT:    [[X:%.*]] = alloca float, align 4, addrspace(5)
+// CHECK-NEXT:    [[Y:%.*]] = alloca float, align 4, addrspace(5)
+// CHECK-NEXT:    [[RETVAL_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL]] to ptr
+// CHECK-NEXT:    [[A_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[A_ADDR]] to ptr
+// CHECK-NEXT:    [[B_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[B_ADDR]] to ptr
+// CHECK-NEXT:    [[C_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[C_ADDR]] to ptr
+// CHECK-NEXT:    [[S_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[S_ADDR]] to ptr
+// CHECK-NEXT:    [[S_INT_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[S_INT_ADDR]] to ptr
+// CHECK-NEXT:    [[X_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[X]] to ptr
+// CHECK-NEXT:    [[Y_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[Y]] to ptr
+// CHECK-NEXT:    store <2 x half> [[A]], ptr [[A_ADDR_ASCAST]], align 4
+// CHECK-NEXT:    store <2 x half> [[B]], ptr [[B_ADDR_ASCAST]], align 4
+// CHECK-NEXT:    store float [[C]], ptr [[C_ADDR_ASCAST]], align 4
+// CHECK-NEXT:    [[FROMBOOL:%.*]] = zext i1 [[S]] to i8
+// CHECK-NEXT:    store i8 [[FROMBOOL]], ptr [[S_ADDR_ASCAST]], align 1
+// CHECK-NEXT:    store i32 [[S_INT]], ptr [[S_INT_ADDR_ASCAST]], align 4
+// CHECK-NEXT:    [[TMP0:%.*]] = load <2 x half>, ptr [[A_ADDR_ASCAST]], align 4
+// CHECK-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr [[B_ADDR_ASCAST]], align 4
+// CHECK-NEXT:    [[TMP2:%.*]] = load float, ptr [[C_ADDR_ASCAST]], align 4
+// CHECK-NEXT:    [[TMP3:%.*]] = load i8, ptr [[S_ADDR_ASCAST]], align 1
+// CHECK-NEXT:    [[TOBOOL:%.*]] = trunc i8 [[TMP3]] to i1
+// CHECK-NEXT:    [[CALL:%.*]] = call float @__ockl_fdot2(<2 x half> [[TMP0]], <2 x half> [[TMP1]], float [[TMP2]], i1 zeroext [[TOBOOL]]) #[[ATTR4]]
+// CHECK-NEXT:    store float [[CALL]], ptr [[X_ASCAST]], align 4
+// CHECK-NEXT:    [[TMP4:%.*]] = load <2 x half>, ptr [[A_ADDR_ASCAST]], align 4
+// CHECK-NEXT:    [[TMP5:%.*]] = load <2 x half>, ptr [[B_ADDR_ASCAST]], align 4
+// CHECK-NEXT:    [[TMP6:%.*]] = load float, ptr [[C_ADDR_ASCAST]], align 4
+// CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr [[S_INT_ADDR_ASCAST]], align 4
+// CHECK-NEXT:    [[TOBOOL1:%.*]] = icmp ne i32 [[TMP7]], 0
+// CHECK-NEXT:    [[CALL2:%.*]] = call float @__ockl_fdot2(<2 x half> [[TMP4]], <2 x half> [[TMP5]], float [[TMP6]], i1 zeroext [[TOBOOL1]]) #[[ATTR4]]
+// CHECK-NEXT:    store float [[CALL2]], ptr [[Y_ASCAST]], align 4
+// CHECK-NEXT:    [[TMP8:%.*]] = load float, ptr [[X_ASCAST]], align 4
+// CHECK-NEXT:    [[TMP9:%.*]] = load float, ptr [[Y_ASCAST]], align 4
+// CHECK-NEXT:    [[ADD:%.*]] = fadd float [[TMP8]], [[TMP9]]
+// CHECK-NEXT:    ret float [[ADD]]
+//
+TEST_FUNC_ATTRS float test_ockl_fdot2(__2f16 a, __2f16 b, float c, bool s, int s_int) {
+  float x = __ockl_fdot2(a, b, c, s);
+  float y = __ockl_fdot2(a, b, c, s_int);
+  return x + y;
+}
+
+
+#ifdef _OPENMP
+#pragma omp end declare target
+#endif
+//// NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+// HIP: {{.*}}
+// OPENMP-C: {{.*}}
+// OPENMP-CPP: {{.*}}