Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

OpenCL reduce API overhead #2081

Merged
merged 1 commit into from
Nov 21, 2018
Merged

OpenCL reduce API overhead #2081

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
59 changes: 11 additions & 48 deletions xmrstak/backend/amd/amd_gpu/gpu.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1116,6 +1116,12 @@ size_t XMRSetJob(GpuContext* ctx, uint8_t* input, size_t input_len, uint64_t tar
printer::inst()->print_msg(L1,"Error %s when calling clSetKernelArg for kernel %d, argument %d.", err_to_str(ret), i + 3, 3);
return ERR_OCL_API;
}

if((clSetKernelArg(ctx->Kernels[kernel_storage][i + 3], 4, sizeof(cl_uint), &numThreads)) != CL_SUCCESS)
{
printer::inst()->print_msg(L1,"Error %s when calling clSetKernelArg for kernel %d, argument %d.", err_to_str(ret), i + 3, 4);
return(ERR_OCL_API);
}
}

return ERR_SUCCESS;
Expand Down Expand Up @@ -1158,8 +1164,6 @@ size_t XMRRunJob(GpuContext* ctx, cl_uint* HashOutput, xmrstak_algo miner_algo)
return ERR_OCL_API;
}

clFinish(ctx->CommandQueues);

size_t Nonce[2] = {ctx->Nonce, 1}, gthreads[2] = { g_thd, 8 }, lthreads[2] = { 8, 8 };
if((ret = clEnqueueNDRangeKernel(ctx->CommandQueues, ctx->Kernels[kernel_storage][0], 2, Nonce, gthreads, lthreads, 0, NULL, NULL)) != CL_SUCCESS)
{
Expand All @@ -1181,64 +1185,23 @@ size_t XMRRunJob(GpuContext* ctx, cl_uint* HashOutput, xmrstak_algo miner_algo)
return ERR_OCL_API;
}

if((ret = clEnqueueReadBuffer(ctx->CommandQueues, ctx->ExtraBuffers[2], CL_FALSE, sizeof(cl_uint) * g_intensity, sizeof(cl_uint), BranchNonces, 0, NULL, NULL)) != CL_SUCCESS)
{
printer::inst()->print_msg(L1,"Error %s when calling clEnqueueReadBuffer to fetch results.", err_to_str(ret));
return ERR_OCL_API;
}

if((ret = clEnqueueReadBuffer(ctx->CommandQueues, ctx->ExtraBuffers[3], CL_FALSE, sizeof(cl_uint) * g_intensity, sizeof(cl_uint), BranchNonces + 1, 0, NULL, NULL)) != CL_SUCCESS)
{
printer::inst()->print_msg(L1,"Error %s when calling clEnqueueReadBuffer to fetch results.", err_to_str(ret));
return ERR_OCL_API;
}

if((ret = clEnqueueReadBuffer(ctx->CommandQueues, ctx->ExtraBuffers[4], CL_FALSE, sizeof(cl_uint) * g_intensity, sizeof(cl_uint), BranchNonces + 2, 0, NULL, NULL)) != CL_SUCCESS)
{
printer::inst()->print_msg(L1,"Error %s when calling clEnqueueReadBuffer to fetch results.", err_to_str(ret));
return ERR_OCL_API;
}

if((ret = clEnqueueReadBuffer(ctx->CommandQueues, ctx->ExtraBuffers[5], CL_FALSE, sizeof(cl_uint) * g_intensity, sizeof(cl_uint), BranchNonces + 3, 0, NULL, NULL)) != CL_SUCCESS)
{
printer::inst()->print_msg(L1,"Error %s when calling clEnqueueReadBuffer to fetch results.", err_to_str(ret));
return ERR_OCL_API;
}

clFinish(ctx->CommandQueues);

for(int i = 0; i < 4; ++i)
{
if(BranchNonces[i])
size_t tmpNonce = ctx->Nonce;
if((ret = clEnqueueNDRangeKernel(ctx->CommandQueues, ctx->Kernels[kernel_storage][i + 3], 1, &tmpNonce, &g_thd, &w_size, 0, NULL, NULL)) != CL_SUCCESS)
{
// Threads
cl_uint numThreads = BranchNonces[i];
if((clSetKernelArg(ctx->Kernels[kernel_storage][i + 3], 4, sizeof(cl_uint), &numThreads)) != CL_SUCCESS)
{
printer::inst()->print_msg(L1,"Error %s when calling clSetKernelArg for kernel %d, argument %d.", err_to_str(ret), i + 3, 4);
return(ERR_OCL_API);
}

// round up to next multiple of w_size
BranchNonces[i] = ((BranchNonces[i] + w_size - 1u) / w_size) * w_size;
// number of global threads must be a multiple of the work group size (w_size)
assert(BranchNonces[i]%w_size == 0);
size_t tmpNonce = ctx->Nonce;
if((ret = clEnqueueNDRangeKernel(ctx->CommandQueues, ctx->Kernels[kernel_storage][i + 3], 1, &tmpNonce, BranchNonces + i, &w_size, 0, NULL, NULL)) != CL_SUCCESS)
{
printer::inst()->print_msg(L1,"Error %s when calling clEnqueueNDRangeKernel for kernel %d.", err_to_str(ret), i + 3);
return ERR_OCL_API;
}
printer::inst()->print_msg(L1,"Error %s when calling clEnqueueNDRangeKernel for kernel %d.", err_to_str(ret), i + 3);
return ERR_OCL_API;
}
}

// this call is blocking therefore the access to the results without cl_finish is fine
if((ret = clEnqueueReadBuffer(ctx->CommandQueues, ctx->OutputBuffer, CL_TRUE, 0, sizeof(cl_uint) * 0x100, HashOutput, 0, NULL, NULL)) != CL_SUCCESS)
{
printer::inst()->print_msg(L1,"Error %s when calling clEnqueueReadBuffer to fetch results.", err_to_str(ret));
return ERR_OCL_API;
}

clFinish(ctx->CommandQueues);
auto & numHashValues = HashOutput[0xFF];
// avoid out of memory read, we have only storage for 0xFF results
if(numHashValues > 0xFF)
Expand Down
24 changes: 12 additions & 12 deletions xmrstak/backend/amd/amd_gpu/opencl/cryptonight.cl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -401,7 +401,7 @@ __kernel void JOIN(cn0,ALGO)(__global ulong *input, __global uint4 *Scratchpad,
AES2[i] = rotate(tmp, 16U);
AES3[i] = rotate(tmp, 24U);
}

__local ulong State_buf[8 * 25];

barrier(CLK_LOCAL_MEM_FENCE);
Expand Down Expand Up @@ -474,12 +474,12 @@ __kernel void JOIN(cn0,ALGO)(__global ulong *input, __global uint4 *Scratchpad,
}

mem_fence(CLK_LOCAL_MEM_FENCE);

// cryptonight_heavy || cryptonight_haven || cryptonight_bittube2
#if (ALGO == 4 || ALGO == 9 || ALGO == 10)
__local uint4 xin[8][8];
{


/* Also left over threads perform this loop.
* The left over thread results will be ignored
Expand Down Expand Up @@ -530,7 +530,7 @@ __kernel void JOIN(cn0,ALGO)(__global ulong *input, __global uint4 *Scratchpad,

)==="
R"===(

// cryptonight_monero_v8 && NVIDIA
#if(ALGO==11 && defined(__NV_CL_C_VERSION))
# define SCRATCHPAD_CHUNK(N) (*(__local uint4*)((__local uchar*)(scratchpad_line) + (idxS ^ (N << 4))))
Expand Down Expand Up @@ -630,7 +630,7 @@ __kernel void JOIN(cn1,ALGO) (__global uint4 *Scratchpad, __global ulong *states
tweak1_2 ^= as_uint2(states[24]);
#endif
}

mem_fence(CLK_LOCAL_MEM_FENCE);

#if(COMP_MODE==1)
Expand Down Expand Up @@ -755,7 +755,7 @@ __kernel void JOIN(cn1,ALGO) (__global uint4 *Scratchpad, __global ulong *states
#endif

((uint4 *)a)[0] ^= tmp;

// cryptonight_monero_v8
#if (ALGO == 11)
# if defined(__NV_CL_C_VERSION)
Expand Down Expand Up @@ -815,7 +815,7 @@ __kernel void JOIN(cn2,ALGO) (__global uint4 *Scratchpad, __global ulong *states
__local uint4 xin1[8][8];
__local uint4 xin2[8][8];
#endif

#if(COMP_MODE==1)
// do not use early return here
if(gIdx < Threads)
Expand Down Expand Up @@ -971,7 +971,7 @@ __kernel void Skein(__global ulong *states, __global uint *BranchBuf, __global u
const ulong idx = get_global_id(0) - get_global_offset(0);

// do not use early return here
if(idx < Threads)
if(idx < BranchBuf[Threads])
{
states += 25 * BranchBuf[idx];

Expand Down Expand Up @@ -1019,8 +1019,8 @@ __kernel void Skein(__global ulong *states, __global uint *BranchBuf, __global u
ulong outIdx = atomic_inc(output + 0xFF);
if(outIdx < 0xFF)
output[outIdx] = BranchBuf[idx] + (uint)get_global_offset(0);
}
}
}
mem_fence(CLK_GLOBAL_MEM_FENCE);
}

Expand Down Expand Up @@ -1052,7 +1052,7 @@ __kernel void JH(__global ulong *states, __global uint *BranchBuf, __global uint
const uint idx = get_global_id(0) - get_global_offset(0);

// do not use early return here
if(idx < Threads)
if(idx < BranchBuf[Threads])
{
states += 25 * BranchBuf[idx];

Expand Down Expand Up @@ -1106,7 +1106,7 @@ __kernel void Blake(__global ulong *states, __global uint *BranchBuf, __global u
const uint idx = get_global_id(0) - get_global_offset(0);

// do not use early return here
if(idx < Threads)
if(idx < BranchBuf[Threads])
{
states += 25 * BranchBuf[idx];

Expand Down Expand Up @@ -1182,7 +1182,7 @@ __kernel void Groestl(__global ulong *states, __global uint *BranchBuf, __global
const uint idx = get_global_id(0) - get_global_offset(0);

// do not use early return here
if(idx < Threads)
if(idx < BranchBuf[Threads])
{
states += 25 * BranchBuf[idx];

Expand Down