diff options
| author | fireice-uk <fireice-uk@users.noreply.github.com> | 2018-02-19 14:10:36 +0000 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2018-02-19 14:10:36 +0000 |
| commit | f785481bb7c1fb887a65a0b19c3e453904bb5474 (patch) | |
| tree | f0ac2fc281c47ce4965189f75e29e7b71002bc85 /xmrstak/backend/amd/amd_gpu/gpu.cpp | |
| parent | 84febdf63edc3cf702ae6747c8b071d5302249de (diff) | |
| parent | 737185ee82bae05953680b1f4c4cdf8646c51b5a (diff) | |
| download | xmr-stak-f785481bb7c1fb887a65a0b19c3e453904bb5474.zip xmr-stak-f785481bb7c1fb887a65a0b19c3e453904bb5474.tar.gz | |
Merge pull request #1087 from psychocrypt/topic-blockedStride3
AMD: option `mem_chunk`and new `strided_index`
Diffstat (limited to 'xmrstak/backend/amd/amd_gpu/gpu.cpp')
| -rw-r--r-- | xmrstak/backend/amd/amd_gpu/gpu.cpp | 11 |
1 files changed, 9 insertions, 2 deletions
diff --git a/xmrstak/backend/amd/amd_gpu/gpu.cpp b/xmrstak/backend/amd/amd_gpu/gpu.cpp index c39c567..054ffc4 100644 --- a/xmrstak/backend/amd/amd_gpu/gpu.cpp +++ b/xmrstak/backend/amd/amd_gpu/gpu.cpp @@ -332,8 +332,8 @@ size_t InitOpenCLGpu(cl_context opencl_ctx, GpuContext* ctx, const char* source_ char options[256]; snprintf(options, sizeof(options), - "-DITERATIONS=%d -DMASK=%d -DWORKSIZE=%llu -DSTRIDED_INDEX=%d", - hasIterations, threadMemMask, int_port(ctx->workSize), ctx->stridedIndex ? 1 : 0); + "-DITERATIONS=%d -DMASK=%d -DWORKSIZE=%llu -DSTRIDED_INDEX=%d -DMEM_CHUNK=%d", + hasIterations, threadMemMask, int_port(ctx->workSize), ctx->stridedIndex, int(1u<<ctx->memChunk)); ret = clBuildProgram(ctx->Program, 1, &ctx->DeviceID, options, NULL, NULL); if(ret != CL_SUCCESS) { @@ -696,6 +696,13 @@ size_t InitOpenCL(GpuContext* ctx, size_t num_gpus, size_t platform_idx) for(int i = 0; i < num_gpus; ++i) { + if(ctx[i].stridedIndex == 2 && (ctx[i].rawIntensity % ctx[i].workSize) != 0) + { + size_t reduced_intensity = (ctx[i].rawIntensity / ctx[i].workSize) * ctx[i].workSize; + ctx[i].rawIntensity = reduced_intensity; + printer::inst()->print_msg(L0, "WARNING AMD: gpu %d intensity is not a multiple of 'worksize', auto reduce intensity to %d", ctx[i].deviceIdx, int(reduced_intensity)); + } + if((ret = InitOpenCLGpu(opencl_ctx, &ctx[i], source_code.c_str())) != ERR_SUCCESS) { return ret; |
