Media/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-09-15 13:01:45 +08:00
Code Issues Packages Projects Releases Wiki Activity

Utilize shortened version of optimal HW thread count in nx4 scenarios.

Browse Source 
- also clean early return if simd size = 0

Change-Id: I9b01df091ab6dd6a3066d1a8762c7fb1530c2804
This commit is contained in:
Mrozek, Michal
2018-01-16 14:27:08 +01:00
parent 92ebb0c3c1
commit dd601ff73a
2 changed files with 84 additions and 71 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
unit_tests/command_queue/local_work_size_tests.cpp
View File

@ -58,13 +58,6 @@ TEST(localWorkSizeTest, given1DimWorkGroupAndSimdEqual8WhenComputeCalledThenLoca
EXPECT_EQ(workGroupSize[0], 9u);
EXPECT_EQ(workGroupSize[1], 1u);
EXPECT_EQ(workGroupSize[2], 1u);
workGroup[0] = 128;
wsInfo.simdSize = 0;
OCLRT::computeWorkgroupSizeND(wsInfo, workGroupSize, workGroup, workDim);
EXPECT_EQ(workGroupSize[0], 1u);
EXPECT_EQ(workGroupSize[1], 1u);
EXPECT_EQ(workGroupSize[2], 1u);
}
TEST(localWorkSizeTest, given1DimWorkGroupAndSimdEqual32WhenComputeCalledThenLocalGroupComputed) {
@ -317,6 +310,27 @@ TEST(localWorkSizeTest, givenKernelWithTileYImagesAndNoBarriersWhenWorkgroupSize
EXPECT_EQ(workGroupSize[2], 1u);
}
TEST(localWorkSizeTest, givenSimd16KernelWithTileYImagesAndNoBarriersWhenWorkgroupSizeIsComputedThenItMimicsTilingPattern) {
WorkSizeInfo wsInfo(256, false, 16, 0u, platformDevices[0]->pPlatform->eRenderCoreFamily, 32u, 0u, true, true);
uint32_t workDim = 2;
size_t workGroup[3] = {1, 1, 1};
size_t workGroupSize[3];
workGroup[0] = 2048;
workGroup[1] = 2048;
OCLRT::computeWorkgroupSizeND(wsInfo, workGroupSize, workGroup, workDim);
EXPECT_EQ(workGroupSize[0], 32u);
EXPECT_EQ(workGroupSize[1], 4u);
EXPECT_EQ(workGroupSize[2], 1u);
workGroup[0] = 1920;
workGroup[1] = 1080;
OCLRT::computeWorkgroupSizeND(wsInfo, workGroupSize, workGroup, workDim);
EXPECT_EQ(workGroupSize[0], 32u);
EXPECT_EQ(workGroupSize[1], 4u);
EXPECT_EQ(workGroupSize[2], 1u);
}
TEST(localWorkSizeTest, givenKernelWithTwoDimensionalGlobalSizesWhenLwsIsComputedThenItHasMaxWorkgroupSize) {
WorkSizeInfo wsInfo(256, 0u, 32, 0u, platformDevices[0]->pPlatform->eRenderCoreFamily, 32u, 0u, false, false);
uint32_t workDim = 2;