thek0tyara/llama-cpp-turboquant
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 4

Reformatted

Browse source
This commit is contained in:
Alejandro Saucedo 2020-10-17 11:19:25 +01:00
parent 872a0bc717
commit 9a64339e95
8 changed files with 95 additions and 72 deletions
Download patch file Download diff file Expand all files Collapse all files

19
test/TestAsyncOperations.cpp Executable file → Normal file
View file

@ -72,13 +72,14 @@ TEST(TestAsyncOperations, TestManagerAsync)
for (uint32_t i = 0; i < numParallel; i++) {
mgr.evalOpDefault<kp::OpAlgoBase<>>(
{ inputsSyncA[i], inputsSyncB[i] },
{ inputsSyncA[i], inputsSyncB[i] },
std::vector<char>(shader.begin(), shader.end()));
}
auto endSync = std::chrono::high_resolution_clock::now();
auto durationSync = std::chrono::duration_cast<std::chrono::microseconds>(endSync - startSync).count();
auto durationSync =
std::chrono::duration_cast<std::chrono::microseconds>(endSync - startSync)
.count();
mgr.evalOpDefault<kp::OpTensorSyncLocal>(inputsSyncB);
@ -86,7 +87,7 @@ TEST(TestAsyncOperations, TestManagerAsync)
EXPECT_EQ(inputsSyncB[i]->data(), resultSync);
}
kp::Manager mgrAsync(0, {0, 2});
kp::Manager mgrAsync(0, { 0, 2 });
std::vector<std::shared_ptr<kp::Tensor>> inputsAsyncA;
std::vector<std::shared_ptr<kp::Tensor>> inputsAsyncB;
@ -107,8 +108,8 @@ TEST(TestAsyncOperations, TestManagerAsync)
for (uint32_t i = 0; i < numParallel; i++) {
mgrAsync.evalOpAsync<kp::OpAlgoBase<>>(
{ inputsAsyncA[i], inputsAsyncB[i] },
"async" + std::to_string(i),
{ inputsAsyncA[i], inputsAsyncB[i] },
"async" + std::to_string(i),
std::vector<char>(shader.begin(), shader.end()));
}
@ -117,7 +118,9 @@ TEST(TestAsyncOperations, TestManagerAsync)
}
auto endAsync = std::chrono::high_resolution_clock::now();
auto durationAsync = std::chrono::duration_cast<std::chrono::microseconds>(endAsync - startAsync).count();
auto durationAsync = std::chrono::duration_cast<std::chrono::microseconds>(
endAsync - startAsync)
.count();
mgrAsync.evalOpDefault<kp::OpTensorSyncLocal>({ inputsAsyncB });
@ -128,6 +131,6 @@ TEST(TestAsyncOperations, TestManagerAsync)
SPDLOG_ERROR("sync {}", durationSync);
SPDLOG_ERROR("async {}", durationAsync);
// The speedup should be at least 40%
// The speedup should be at least 40%
EXPECT_LT(durationAsync, durationSync * 0.6);
}

12
test/TestMultipleAlgoExecutions.cpp
View file

@ -258,10 +258,10 @@ TEST(TestMultipleAlgoExecutions, ManagerEvalMultSourceStrOpCreate)
)");
mgr.evalOpDefault<kp::OpAlgoBase<>>(
{ tensorInA, tensorInB, tensorOut },
std::vector<char>(shader.begin(), shader.end()));
{ tensorInA, tensorInB, tensorOut },
std::vector<char>(shader.begin(), shader.end()));
mgr.evalOpDefault<kp::OpTensorSyncLocal>({tensorOut});
mgr.evalOpDefault<kp::OpTensorSyncLocal>({ tensorOut });
EXPECT_EQ(tensorOut->data(), std::vector<float>({ 0.0, 4.0, 12.0 }));
}
@ -295,10 +295,10 @@ TEST(TestMultipleAlgoExecutions, ManagerEvalMultSourceStrMgrCreate)
)");
mgr.evalOpDefault<kp::OpAlgoBase<>>(
{ tensorInA, tensorInB, tensorOut },
std::vector<char>(shader.begin(), shader.end()));
{ tensorInA, tensorInB, tensorOut },
std::vector<char>(shader.begin(), shader.end()));
mgr.evalOpDefault<kp::OpTensorSyncLocal>({tensorOut});
mgr.evalOpDefault<kp::OpTensorSyncLocal>({ tensorOut });
EXPECT_EQ(tensorOut->data(), std::vector<float>({ 0.0, 4.0, 12.0 }));
}