Details
Details
Diff Detail
Diff Detail
- Repository
- rL LLVM
Event Timeline
MicroBenchmarks/ImageProcessing/Blur/blur.h | ||
---|---|---|
15–18 ↗ | (On Diff #155554) | Do these functions need the array sizes at all? Why not just use int **? |
MicroBenchmarks/ImageProcessing/Blur/main.cpp | ||
167 ↗ | (On Diff #155554) | You probably want a ::benchmark::DoNotOptimize(...) around this and other calls to the kernel. |
MicroBenchmarks/ImageProcessing/Blur/main.cpp | ||
---|---|---|
167 ↗ | (On Diff #155554) | I don't think I can use "::benchmark::DoNotOptimize(...)" here as the functions return void and DoNotOptimize forces the compiler to flush pending writes to memory. |
MicroBenchmarks/ImageProcessing/Blur/main.cpp | ||
---|---|---|
175–177 ↗ | (On Diff #155760) | It may be. |
MicroBenchmarks/ImageProcessing/Blur/boxBlurKernel.cpp | ||
---|---|---|
10–11 ↗ | (On Diff #155760) | Can you change inpImage/outImage to use C99 array parameter syntax? |
MicroBenchmarks/ImageProcessing/Blur/main.cpp | ||
175–177 ↗ | (On Diff #155760) | I suggest to leave it as is for the moment. If we found a canonical approach, we can change this and the other benchmarks afterwards. Could you add a comment that this is supposed keep the compiler to optimize the computation away? |