Skip to content

opencl: fix warnings and clean up profiling #16688

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

Merged
merged 2 commits into from
Oct 21, 2025
+9 −16
Merged

opencl: fix warnings and clean up profiling #16688

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
a1ceb3d
opencl: remove unused headers, fix warnings
lhez Oct 20, 2025
9be4b58
opencl: clean up profiling, only keep kernel time
lhez Oct 20, 2025
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
25 changes: 9 additions & 16 deletions ggml/src/ggml-opencl/ggml-opencl.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,13 +15,12 @@

#include <CL/cl.h>

#include <inttypes.h>
#include <string.h>

#include <cstddef>
#include <cstdint>
#include <atomic>
#include <fstream>
#include <limits>
#include <vector>
#include <string>
#include <cmath>
Expand Down Expand Up @@ -533,25 +532,17 @@ struct ggml_backend_opencl_context {
}

// Dump a csv
float total_kernel_time = 0;
fprintf(fperf, "op name, kernel name, queued duration (ms), submit duration(ms), exec duration (ms), complete duration (ms), total duration (ms), global size, local size, output size\n");
fprintf(fperf, "op name, kernel name, exec duration (ms), global size, local size, output size\n");
for (const ProfilingInfo & info : profiling_info) {
total_kernel_time += info.cmd_duration_ns/1.e6f;
fprintf(fperf, "%s,%s,%f,%f,%f,%f,%f,%zux%zux%zu,%zux%zux%zu,%zux%zux%zux%zu\n",
fprintf(fperf, "%s,%s,%f,%zux%zux%zu,%zux%zux%zu,%zux%zux%zux%zu\n",
info.op_name.c_str(), info.kernel_name.c_str(),
info.cmd_queued_duration_ns/1.e6f,
info.cmd_submit_duration_ns/1.e6f,
info.cmd_duration_ns/1.e6f,
info.cmd_complete_duration_ns/1.e6f,
info.cmd_total_duration_ns/1.e6f,
info.global_size[0], info.global_size[1], info.global_size[2],
info.local_size[0], info.local_size[1], info.local_size[2],
info.output_size[0], info.output_size[1], info.output_size[2], info.output_size[3]);
}
fclose(fperf);

GGML_LOG_INFO("ggml_opencl: total kernel time: %f\n", total_kernel_time);

// Dump a simple chrome trace
FILE* ftrace = fopen("cl_trace.json", "w");
if (!ftrace) {
Expand All @@ -561,14 +552,14 @@ struct ggml_backend_opencl_context {

fprintf(ftrace, "[\n");
for (const ProfilingInfo & info : profiling_info) {
fprintf(ftrace, "{\"name\": \"%s\", \"cat\": \"OpenCL\", \"ph\": \"B\", \"ts\": %llu, \"pid\": \"\", \"tid\": \"Host\"},\n",
fprintf(ftrace, "{\"name\": \"%s\", \"cat\": \"OpenCL\", \"ph\": \"B\", \"ts\": %" PRIu64 ", \"pid\": \"\", \"tid\": \"Host\"},\n",
info.kernel_name.c_str(), info.cmd_queued/1000);
fprintf(ftrace, "{\"name\": \"%s\", \"cat\": \"OpenCL\", \"ph\": \"E\", \"ts\": %llu, \"pid\": \"\", \"tid\": \"Host\"},\n",
fprintf(ftrace, "{\"name\": \"%s\", \"cat\": \"OpenCL\", \"ph\": \"E\", \"ts\": %" PRIu64 ", \"pid\": \"\", \"tid\": \"Host\"},\n",
info.kernel_name.c_str(), info.cmd_submit/1000);

fprintf(ftrace, "{\"name\": \"%s\", \"cat\": \"OpenCL\", \"ph\": \"B\", \"ts\": %llu, \"pid\": \"\", \"tid\": \"Device\"},\n",
fprintf(ftrace, "{\"name\": \"%s\", \"cat\": \"OpenCL\", \"ph\": \"B\", \"ts\": %" PRIu64 ", \"pid\": \"\", \"tid\": \"Device\"},\n",
info.kernel_name.c_str(), info.cmd_start/1000);
fprintf(ftrace, "{\"name\": \"%s\", \"cat\": \"OpenCL\", \"ph\": \"E\", \"ts\": %llu, \"pid\": \"\", \"tid\": \"Device\"},\n",
fprintf(ftrace, "{\"name\": \"%s\", \"cat\": \"OpenCL\", \"ph\": \"E\", \"ts\": %" PRIu64 ", \"pid\": \"\", \"tid\": \"Device\"},\n",
info.kernel_name.c_str(), info.cmd_end/1000);
}
fclose(ftrace);
Expand Down Expand Up @@ -7652,6 +7643,8 @@ static void ggml_cl_mul_mat_id(ggml_backend_t backend, const ggml_tensor * src0,
const cl_ulong nb21 = src2->nb[1];
const cl_ulong nb20 = src2->nb[0];

UNUSED(nb20);

const int ne0 = dst->ne[0];
const int ne1 = dst->ne[1];

Expand Down
Loading