AI/rocm-systems
2
0
Fork 0
Código Issues Pull Requests Ações Pacotes Projetos Versões Wiki Atividade

Missing commit.

Ver Código-Fonte
Esse commit está contido em:
Alex Voicu
2018-05-23 17:57:09 +01:00
pai 7042fe6067
commit ecefdd6541
7 arquivos alterados com 2128 adições e 831 exclusões
Baixar Arquivo de Patch Baixar Arquivo de Diff Expandir todos os arquivos Colapsar todos os arquivos
hipamd/tests/src/deviceLib/hipTestHalf.cpp
+32 -16
Ver Arquivo
@@ -34,7 +34,7 @@ THE SOFTWARE.
#if __HIP_ARCH_GFX803__ || __HIP_ARCH_GFX900__ || __HIP_ARCH_GFX906__
__global__ void __halfMath(hipLaunchParm lp, __half* A, __half* B, __half* C) {
__global__ void __halfMath(__half* A, __half* B, __half* C) {
int tx = threadIdx.x;
__half a = A[tx];
__half b = B[tx];
@@ -47,10 +47,10 @@ __global__ void __halfMath(hipLaunchParm lp, __half* A, __half* B, __half* C) {
c = __hsub_sat(b, c);
c = __hmul(a, c);
c = __hmul_sat(b, c);
c = hdiv(a, c);
c = __hdiv(a, c);
}
__global__ void __half2Math(hipLaunchParm lp, __half2* A, __half2* B, __half2* C) {
__global__ void __half2Math(__half2* A, __half2* B, __half2* C) {
int tx = threadIdx.x;
__half2 a = A[tx];
__half2 b = B[tx];
@@ -65,12 +65,12 @@ __global__ void __half2Math(hipLaunchParm lp, __half2* A, __half2* B, __half2* C
c = __hmul2_sat(b, c);
}
__global__ void kernel_hisnan(hipLaunchParm lp, __half* input, int* output) {
__global__ void kernel_hisnan(__half* input, int* output) {
int tx = threadIdx.x;
output[tx] = __hisnan(input[tx]);
}
__global__ void kernel_hisinf(hipLaunchParm lp, __half* input, int* output) {
__global__ void kernel_hisinf(__half* input, int* output) {
int tx = threadIdx.x;
output[tx] = __hisinf(input[tx]);
}
@@ -93,7 +93,8 @@ void check_hisnan(int NUM_INPUTS, __half* inputCPU, __half* inputGPU) {
hipMalloc((void**)&outputGPU, memsize);
// launch the kernel
hipLaunchKernel(kernel_hisnan, dim3(1), dim3(NUM_INPUTS), 0, 0, inputGPU, outputGPU);
hipLaunchKernelGGL(
kernel_hisnan, dim3(1), dim3(NUM_INPUTS), 0, 0, inputGPU, outputGPU);
// copy output from device
int* outputCPU = (int*) malloc(memsize);
@@ -103,12 +104,18 @@ void check_hisnan(int NUM_INPUTS, __half* inputCPU, __half* inputGPU) {
for (int i=0; i<NUM_INPUTS; i++) {
if ((2 <= i) && (i <= 5)) { // inputs are nan, output should be true
if (outputCPU[i] == 0) {
failed("__hisnan() returned false for %f (input idx = %d)\n", inputCPU[i], i);
failed(
"__hisnan() returned false for %f (input idx = %d)\n",
static_cast<float>(inputCPU[i]),
i);
}
}
else { // inputs are NOT nan, output should be false
if (outputCPU[i] != 0) {
failed("__hisnan() returned true for %f (input idx = %d)\n", inputCPU[i], i);
failed(
"__hisnan() returned true for %f (input idx = %d)\n",
static_cast<float>(inputCPU[i]),
i);
}
}
}
@@ -129,7 +136,8 @@ void check_hisinf(int NUM_INPUTS, __half* inputCPU, __half* inputGPU) {
hipMalloc((void**)&outputGPU, memsize);
// launch the kernel
hipLaunchKernel(kernel_hisinf, dim3(1), dim3(NUM_INPUTS), 0, 0, inputGPU, outputGPU);
hipLaunchKernelGGL(
kernel_hisinf, dim3(1), dim3(NUM_INPUTS), 0, 0, inputGPU, outputGPU);
// copy output from device
int* outputCPU = (int*) malloc(memsize);
@@ -139,12 +147,18 @@ void check_hisinf(int NUM_INPUTS, __half* inputCPU, __half* inputGPU) {
for (int i=0; i<NUM_INPUTS; i++) {
if ((0 <= i) && (i <= 1)) { // inputs are inf, output should be true
if (outputCPU[i] == 0) {
failed("__hisinf() returned false for %f (input idx = %d)\n", inputCPU[i], i);
failed(
"__hisinf() returned false for %f (input idx = %d)\n",
static_cast<float>(inputCPU[i]),
i);
}
}
else { // inputs are NOT inf, output should be false
if (outputCPU[i] != 0) {
failed("__hisinf() returned true for %f (input idx = %d)\n", inputCPU[i], i);
failed(
"__hisinf() returned true for %f (input idx = %d)\n",
static_cast<float>(inputCPU[i]),
i);
}
}
}
@@ -160,11 +174,11 @@ void check_hisinf(int NUM_INPUTS, __half* inputCPU, __half* inputGPU) {
void checkFunctional() {
// allocate memory
// allocate memory
const int NUM_INPUTS = 16;
auto memsize = NUM_INPUTS * sizeof(__half);
__half* inputCPU = (__half*) malloc(memsize);
// populate inputs
inputCPU[0] = host_ushort_as_half(0x7c00); // inf
inputCPU[1] = host_ushort_as_half(0xfc00); // -inf
@@ -207,7 +221,8 @@ int main() {
hipMalloc(&A, HALF_SIZE);
hipMalloc(&B, HALF_SIZE);
hipMalloc(&C, HALF_SIZE);
hipLaunchKernel(__halfMath, dim3(1, 1, 1), dim3(LEN, 1, 1), 0, 0, A, B, C);
hipLaunchKernelGGL(
__halfMath, dim3(1, 1, 1), dim3(LEN, 1, 1), 0, 0, A, B, C);
hipFree(A);
hipFree(B);
hipFree(C);
@@ -215,13 +230,14 @@ int main() {
hipMalloc(&A2, HALF2_SIZE);
hipMalloc(&B2, HALF2_SIZE);
hipMalloc(&C2, HALF2_SIZE);
hipLaunchKernel(__half2Math, dim3(1, 1, 1), dim3(LEN, 1, 1), 0, 0, A2, B2, C2);
hipLaunchKernelGGL(
__half2Math, dim3(1, 1, 1), dim3(LEN, 1, 1), 0, 0, A2, B2, C2);
hipFree(A2);
hipFree(B2);
hipFree(C2);
// run some functional checks
checkFunctional();
passed();
}
hipamd/tests/src/deviceLib/hipTestNativeHalf.cpp
+142
Ver Arquivo
@@ -0,0 +1,142 @@
/*
Copyright (c) 2015-2017 Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
/* HIT_START
* BUILD: %t %s ../test_common.cpp EXCLUDE_HIP_PLATFORM nvcc
* RUN: %t
* HIT_END
*/
#include <hip/hip_fp16.h>
#include "hip/hip_runtime.h"
#include "test_common.h"
#if __HIP_ARCH_GFX803__ || __HIP_ARCH_GFX900__ || __HIP_ARCH_GFX906__
__global__
__attribute__((optnone))
void __halfTest(bool* result) {
// Construction
__half a{1}; result[0] = __heq(a, 1);
a = __half{1.0f}; result[0] = __heq(a, 1) && result[0];
a = __half{1.0}; result[0] = __heq(a, 1) && result[0];
a = __half{static_cast<unsigned short>(1)};
result[0] = __heq(a, 1) && result[0];
a = __half{static_cast<short>(1)}; result[0] = __heq(a, 1) && result[0];
a = __half{1u}; result[0] = __heq(a, 1) && result[0];
a = __half{1ul}; result[0] = __heq(a, 1) && result[0];
a = __half{1l}; result[0] = __heq(a, 1) && result[0];
a = __half{1ll}; result[0] = __heq(a, 1) && result[0];
a = __half{1ull}; result[0] = __heq(a, 1) && result[0];
// Assignment
a = 0.0f; result[0] = __heq(a, 0) && result[0];
a = 1.0; result[0] = __heq(a, 1) && result[0];
a = __half_raw{2}; result[0] = __heq(a, 2) && result[0];
// Nullary
result[0] = __heq(a, +a) && result[0];
result[0] = __heq(__hneg(a), -a) && result[0];
// Unary arithmetic
result[0] = __heq(a += 0, a) && result[0];
result[0] = __heq(a -= 0, a) && result[0];
result[0] = __heq(a *= 1, a) && result[0];
result[0] = __heq(a /= 1, a) && result[0];
// Binary arithmetic
result[0] = __heq((a + a), __hadd(a, a)) && result[0];
result[0] = __heq((a - a), __hsub(a, a)) && result[0];
result[0] = __heq((a * a), __hmul(a, a)) && result[0];
result[0] = __heq((a / a), __hdiv(a, a)) && result[0];
// Relations
result[0] = (a == a) && result[0];
result[0] = !(a != a) && result[0];
result[0] = (a <= a) && result[0];
result[0] = (a >= a) && result[0];
result[0] = !(a < a) && result[0];
result[0] = !(a > a) && result[0];
}
__device__
bool to_bool(const __half2& x)
{
auto r = static_cast<const __half2_raw&>(x);
return r.data.x != 0 && r.data.y != 0;
}
__global__
__attribute__((optnone))
void __half2Test(bool* result) {
// Construction
__half2 a{1};
result[0] = to_bool(__heq2(a, 1));
a = __half2{__half{1}, __half{1}};
result[0] = to_bool(__heq2(a, {1, 1})) && result[0];
// Assignment
a = __half2_raw{2}; result[0] = to_bool(__heq2(a, {2, 2})) && result[0];
// Nullary
result[0] = to_bool(__heq2(a, +a)) && result[0];
result[0] = to_bool(__heq2(__hneg2(a), -a)) && result[0];
// Unary arithmetic
result[0] = to_bool(__heq2(a += 0, a)) && result[0];
result[0] = to_bool(__heq2(a -= 0, a)) && result[0];
result[0] = to_bool(__heq2(a *= 1, a)) && result[0];
result[0] = to_bool(__heq2(a /= 1, a)) && result[0];
// Binary arithmetic
result[0] = to_bool(__heq2((a + a), __hadd2(a, a))) && result[0];
result[0] = to_bool(__heq2((a - a), __hsub2(a, a))) && result[0];
result[0] = to_bool(__heq2((a * a), __hmul2(a, a))) && result[0];
result[0] = to_bool(__heq2((a / a), __h2div(a, a))) && result[0];
// Relations
result[0] = (a == a) && result[0];
result[0] = !(a != a) && result[0];
result[0] = (a <= a) && result[0];
result[0] = (a >= a) && result[0];
result[0] = !(a < a) && result[0];
result[0] = !(a > a) && result[0];
}
#endif
int main() {
bool* result{nullptr};
hipHostMalloc(&result, 1);
result[0] = false;
hipLaunchKernelGGL(__halfTest, dim3(1, 1, 1), dim3(1, 1, 1), 0, 0, result);
hipDeviceSynchronize();
if (!result[0]) { failed("Failed __half tests."); }
result[0] = false;
hipLaunchKernelGGL(__half2Test, dim3(1, 1, 1), dim3(1, 1, 1), 0, 0, result);
hipDeviceSynchronize();
if (!result[0]) { failed("Failed __half2 tests."); }
passed();
}