AMD: `mem_chunk`and new `strided_index`

- add new option for `strided_index`
- add additional option if `strided_index == 2` to controll the memory chunk with

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;