AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

SWDEV-521760 - Fix and enable disabled HIP tests from cooperative groups group (#2027)

Browse Source 
* Reworked Unit_hipLaunchCooperativeKernel_Basic and Unit_hipLaunchCooperativeKernelMultiDevice_Basic
* Introduce reduction_factor for coop groups tests. Fix Unit_Coalesced_Group_Tiled_Partition_Sync_Positive_Basic
* Fix always false requirement by adding a cast
* Change data type to unsigned long long to align with cuda
* Change literal type to double to ensure proper type casting
* Remove formatting comments
This commit is contained in:
amilanov-amd
2026-01-27 11:51:08 +01:00
committed by GitHub
parent c4a9567492
commit cac67a0f32
11 changed files with 226 additions and 218 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/hip-tests/catch/include/cpu_grid.h
+30 -52
View File
@@ -114,11 +114,9 @@ struct CPUMultiGrid {
};
/* Generate dimensions for 1D, 2D and 3D blocks of threads */
inline dim3 GenerateThreadDimensions() {
inline dim3 GenerateThreadDimensionsImpl(const std::initializer_list<double>& multipliers) {
hipDeviceProp_t props;
HIP_CHECK(hipGetDeviceProperties(&props, 0));
const auto multipliers = {0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0, 1.1, 1.2, 1.3,
1.4, 1.5, 1.6, 1.7, 1.8, 1.9, 2.0, 2.1, 2.2, 2.3, 2.4, 2.5};
return GENERATE_COPY(
dim3(1, 1, 1), dim3(props.maxThreadsDim[0], 1, 1), dim3(1, props.maxThreadsDim[1], 1),
dim3(1, 1, props.maxThreadsDim[2]),
@@ -135,59 +133,39 @@ inline dim3 GenerateThreadDimensions() {
dim3(props.warpSize - 1, 3, 3), dim3(props.warpSize + 1, 3, 3));
}
inline dim3 GenerateThreadDimensions() {
const auto multipliers = {0.1, 0.5, 1.0, 1.5, 2.0};
return GenerateThreadDimensionsImpl(multipliers);
}
inline dim3 GenerateThreadDimensionsForShuffle() {
const auto multipliers = {0.5, 1.0, 2.0};
return GenerateThreadDimensionsImpl(multipliers);
}
/* Generate dimensions for 1D, 2D and 3D grids of blocks */
inline dim3 GenerateBlockDimensionsImpl(const std::initializer_list<double>& multipliers) {
hipDeviceProp_t props;
HIP_CHECK(hipGetDeviceProperties(&props, 0));
return GENERATE_COPY(dim3(1, 1, 1),
map([sm = props.multiProcessorCount](
double i) { return dim3(static_cast<int>(i * sm), 1, 1); },
values(multipliers)),
map([sm = props.multiProcessorCount](
double i) { return dim3(1, static_cast<int>(i * sm), 1); },
values(multipliers)),
map([sm = props.multiProcessorCount](
double i) { return dim3(1, 1, static_cast<int>(i * sm)); },
values(multipliers)),
dim3(5, 5, 5));
}
inline dim3 GenerateBlockDimensions() {
hipDeviceProp_t props;
HIP_CHECK(hipGetDeviceProperties(&props, 0));
const auto multipliers = {0.5, 0.9, 1.0, 1.1, 1.5, 1.9, 2.0, 3.0, 4.0};
return GENERATE_COPY(dim3(1, 1, 1),
map([sm = props.multiProcessorCount](
double i) { return dim3(static_cast<int>(i * sm), 1, 1); },
values(multipliers)),
map([sm = props.multiProcessorCount](
double i) { return dim3(1, static_cast<int>(i * sm), 1); },
values(multipliers)),
map([sm = props.multiProcessorCount](
double i) { return dim3(1, 1, static_cast<int>(i * sm)); },
values(multipliers)),
dim3(5, 5, 5));
const auto multipliers = {0.5, 1.0, 1.5, 2.0};
return GenerateBlockDimensionsImpl(multipliers);
}
/* Generate dimensions for 1D, 2D and 3D blocks of threads - reduced set */
inline dim3 GenerateThreadDimensionsForShuffle() {
hipDeviceProp_t props;
HIP_CHECK(hipGetDeviceProperties(&props, 0));
const auto multipliers = {0.5, 0.9, 1.0, 1.5, 2.0};
return GENERATE_COPY(
dim3(1, 1, 1), dim3(props.maxThreadsDim[0], 1, 1), dim3(1, props.maxThreadsDim[1], 1),
dim3(1, 1, props.maxThreadsDim[2]),
map([max = props.maxThreadsDim[0], warp_size = props.warpSize](
double i) { return dim3(std::min(static_cast<int>(i * warp_size), max), 1, 1); },
values(multipliers)),
map([max = props.maxThreadsDim[1], warp_size = props.warpSize](
double i) { return dim3(1, std::min(static_cast<int>(i * warp_size), max), 1); },
values(multipliers)),
map([max = props.maxThreadsDim[2], warp_size = props.warpSize](
double i) { return dim3(1, 1, std::min(static_cast<int>(i * warp_size), max)); },
values(multipliers)),
dim3(16, 8, 8), dim3(32, 32, 1), dim3(64, 8, 2), dim3(16, 16, 3),
dim3(props.warpSize - 1, 3, 3), dim3(props.warpSize + 1, 3, 3));
}
/* Generate dimensions for 1D, 2D and 3D grids of blocks - reduced set */
inline dim3 GenerateBlockDimensionsForShuffle() {
hipDeviceProp_t props;
HIP_CHECK(hipGetDeviceProperties(&props, 0));
const auto multipliers = {0.5, 1.0};
return GENERATE_COPY(dim3(1, 1, 1),
map([sm = props.multiProcessorCount](
double i) { return dim3(static_cast<int>(i * sm), 1, 1); },
values(multipliers)),
map([sm = props.multiProcessorCount](
double i) { return dim3(1, static_cast<int>(i * sm), 1); },
values(multipliers)),
map([sm = props.multiProcessorCount](
double i) { return dim3(1, 1, static_cast<int>(i * sm)); },
values(multipliers)),
dim3(5, 5, 5));
return GenerateBlockDimensionsImpl(multipliers);
}