compute-runtime/unit_tests/command_queue/local_id_tests.cpp

258 lines
9.1 KiB
C++

/*
* Copyright (c) 2017, Intel Corporation
*
* 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.
*/
#include "runtime/command_queue/local_id_gen.h"
#include "runtime/helpers/aligned_memory.h"
#include "runtime/helpers/ptr_math.h"
#include "gtest/gtest.h"
#include <algorithm>
#include <cstdint>
using namespace OCLRT;
TEST(LocalID, GRFsPerThread_SIMD8) {
uint32_t simd = 8;
EXPECT_EQ(1u, getGRFsPerThread(simd));
}
TEST(LocalID, GRFsPerThread_SIMD16) {
uint32_t simd = 16;
EXPECT_EQ(1u, getGRFsPerThread(simd));
}
TEST(LocalID, GRFsPerThread_SIMD32) {
uint32_t simd = 32;
EXPECT_EQ(2u, getGRFsPerThread(simd));
}
TEST(LocalID, ThreadsPerWorkgroup) {
size_t lws = 33;
uint32_t simd = 32;
EXPECT_EQ(2u, getThreadsPerWG(simd, lws));
}
TEST(LocalID, PerThreadSizeLocalIDs_SIMD8) {
uint32_t simd = 8;
// 3 channels (x,y,z) * 1 GRFs per thread (@SIMD8)
EXPECT_EQ(3 * sizeof(GRF), getPerThreadSizeLocalIDs(simd));
}
TEST(LocalID, PerThreadSizeLocalIDs_SIMD16) {
uint32_t simd = 16;
// 3 channels (x,y,z) * 1 GRFs per thread (@SIMD16)
EXPECT_EQ(3 * sizeof(GRF), getPerThreadSizeLocalIDs(simd));
}
TEST(LocalID, PerThreadSizeLocalIDs_SIMD32) {
uint32_t simd = 32;
// 3 channels (x,y,z) * 2 GRFs per thread (@SIMD32)
EXPECT_EQ(6 * sizeof(GRF), getPerThreadSizeLocalIDs(simd));
}
struct LocalIDFixture : public ::testing::TestWithParam<std::tuple<int, int, int, int>> {
void SetUp() override {
simd = std::get<0>(GetParam());
localWorkSizeX = std::get<1>(GetParam());
localWorkSizeY = std::get<2>(GetParam());
localWorkSizeZ = std::get<3>(GetParam());
localWorkSize = localWorkSizeX * localWorkSizeY * localWorkSizeZ;
if (localWorkSize > 256) {
localWorkSizeY = std::min(256 / localWorkSizeX, localWorkSizeY);
localWorkSizeZ = std::min(256 / (localWorkSizeX * localWorkSizeY), localWorkSizeZ);
localWorkSize = localWorkSizeX * localWorkSizeY * localWorkSizeZ;
}
const auto bufferSize = 32 * 3 * 16 * sizeof(uint16_t);
buffer = reinterpret_cast<uint16_t *>(alignedMalloc(bufferSize, 32));
memset(buffer, 0xff, bufferSize);
}
void TearDown() override {
alignedFree(buffer);
}
void validateIDWithinLimits(uint32_t simd, uint32_t lwsX, uint32_t lwsY, uint32_t lwsZ) {
auto idsPerThread = simd;
// As per BackEnd HLD, SIMD32 has 32 localIDs per channel. SIMD8/16 has up to 16 localIDs.
auto skipPerThread = simd == 32 ? 32 : 16;
auto pBufferX = buffer;
auto pBufferY = pBufferX + skipPerThread;
auto pBufferZ = pBufferY + skipPerThread;
auto numWorkItems = lwsX * lwsY * lwsZ;
size_t itemIndex = 0;
while (numWorkItems > 0) {
EXPECT_LT(pBufferX[itemIndex], lwsX) << simd << " " << lwsX << " " << lwsY << " " << lwsZ;
EXPECT_LT(pBufferY[itemIndex], lwsY) << simd << " " << lwsX << " " << lwsY << " " << lwsZ;
EXPECT_LT(pBufferZ[itemIndex], lwsZ) << simd << " " << lwsX << " " << lwsY << " " << lwsZ;
++itemIndex;
if (idsPerThread == itemIndex) {
pBufferX += skipPerThread * 3;
pBufferY += skipPerThread * 3;
pBufferZ += skipPerThread * 3;
itemIndex = 0;
}
--numWorkItems;
}
}
void validateAllWorkItemsCovered(uint32_t simd, uint32_t lwsX, uint32_t lwsY, uint32_t lwsZ) {
auto idsPerThread = simd;
// As per BackEnd HLD, SIMD32 has 32 localIDs per channel. SIMD8/16 has up to 16 localIDs.
auto skipPerThread = simd == 32 ? 32 : 16;
auto pBufferX = buffer;
auto pBufferY = pBufferX + skipPerThread;
auto pBufferZ = pBufferY + skipPerThread;
auto numWorkItems = lwsX * lwsY * lwsZ;
// Initialize local ID hit table
uint32_t localIDHitTable[8];
memset(localIDHitTable, 0, sizeof(localIDHitTable));
size_t itemIndex = 0;
while (numWorkItems > 0) {
// Flatten out the IDs
auto workItem = pBufferX[itemIndex] + pBufferY[itemIndex] * lwsX + pBufferZ[itemIndex] * lwsX * lwsY;
ASSERT_LT(workItem, 256u);
// Look up in the hit table
auto &hitItem = localIDHitTable[workItem / 32];
auto hitBit = 1 << (workItem % 32);
// No double-hits
EXPECT_EQ(0u, hitItem & hitBit);
// Set that work item as hit
hitItem |= hitBit;
++itemIndex;
if (idsPerThread == itemIndex) {
pBufferX += skipPerThread * 3;
pBufferY += skipPerThread * 3;
pBufferZ += skipPerThread * 3;
itemIndex = 0;
}
--numWorkItems;
}
// All entries in hit table should be in form of n^2 - 1
for (uint32_t i : localIDHitTable) {
EXPECT_EQ(0u, i & (i + 1));
}
}
void dumpBuffer(uint32_t simd, uint32_t lwsX, uint32_t lwsY, uint32_t lwsZ) {
auto workSize = lwsX * lwsY * lwsZ;
auto threads = (workSize + simd - 1) / simd;
auto pBuffer = buffer;
// As per BackEnd HLD, SIMD32 has 32 localIDs per channel. SIMD8/16 has up to 16 localIDs.
auto skipPerThread = simd == 32 ? 32 : 16;
while (threads-- > 0) {
auto lanes = std::min(workSize, simd);
for (auto dimension = 0u; dimension < 3u; ++dimension) {
for (auto lane = 0u; lane < lanes; ++lane) {
printf("%04d ", (unsigned int)pBuffer[lane]);
}
pBuffer += skipPerThread;
printf("\n");
}
workSize -= simd;
}
}
// Test parameters
uint32_t localWorkSizeX;
uint32_t localWorkSizeY;
uint32_t localWorkSizeZ;
uint32_t localWorkSize;
uint32_t simd;
// Provide support for a max LWS of 256
// 32 threads @ SIMD8
// 3 channels (x/y/z)
// 16 lanes per thread (SIMD8 - only 8 used)
uint16_t *buffer;
};
TEST_P(LocalIDFixture, checkIDWithinLimits) {
generateLocalIDs(buffer, simd, localWorkSizeX, localWorkSizeY, localWorkSizeZ);
validateIDWithinLimits(simd, localWorkSizeX, localWorkSizeY, localWorkSizeZ);
}
TEST_P(LocalIDFixture, checkAllWorkItemsCovered) {
generateLocalIDs(buffer, simd, localWorkSizeX, localWorkSizeY, localWorkSizeZ);
validateAllWorkItemsCovered(simd, localWorkSizeX, localWorkSizeY, localWorkSizeZ);
}
TEST_P(LocalIDFixture, sizeCalculationLocalIDs) {
auto workItems = localWorkSizeX * localWorkSizeY * localWorkSizeZ;
auto sizeTotalPerThreadData = getThreadsPerWG(simd, workItems) * getPerThreadSizeLocalIDs(simd);
// Should be multiple of GRFs
auto sizeGRF = sizeof(GRF);
EXPECT_EQ(0u, sizeTotalPerThreadData % sizeGRF);
auto numGRFsPerThread = (simd == 32) ? 2 : 1;
auto numThreadsExpected = (workItems + simd - 1) / simd;
auto numGRFsExpected = 3 * numGRFsPerThread * numThreadsExpected;
EXPECT_EQ(numGRFsExpected * sizeGRF, sizeTotalPerThreadData);
}
#define SIMDParams ::testing::Values(8, 16, 32)
#if HEAVY_DUTY_TESTING
#define LWSXParams ::testing::Values(1, 7, 8, 9, 15, 16, 17, 31, 32, 33, 64, 128, 256)
#define LWSYParams ::testing::Values(1, 2, 3, 4, 5, 6, 7, 8)
#define LWSZParams ::testing::Values(1, 2, 3, 4)
#else
#define LWSXParams ::testing::Values(1, 7, 8, 9, 15, 16, 17, 31, 32, 33, 64, 128, 256)
#define LWSYParams ::testing::Values(1, 2, 4, 8)
#define LWSZParams ::testing::Values(1)
#endif
INSTANTIATE_TEST_CASE_P(AllCombinations, LocalIDFixture, ::testing::Combine(SIMDParams, LWSXParams, LWSYParams, LWSZParams));
// To debug a specific configuration replace the list of Values with specific values.
// NOTE: You'll need a unique test prefix
INSTANTIATE_TEST_CASE_P(SingleTest, LocalIDFixture,
::testing::Combine(
::testing::Values(32), //SIMD
::testing::Values(5), //LWSX
::testing::Values(6), //LWSY
::testing::Values(7))); //LWSZ