vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

wrap all CL calls in checks.

Browse source
This commit is contained in:
Henri Vasserman 2023-05-19 13:08:57 +03:00
parent 772e3fbe12
commit 35dbc8d799
No known key found for this signature in database
GPG key ID: 2995FC0F58B1A986
1 changed files with 71 additions and 87 deletions
Download patch file Download diff file Expand all files Collapse all files

134
ggml-opencl.c
View file

@ -126,11 +126,22 @@ __kernel void dequantize_row_q8_0(__global struct block_q8_0* x, __global float*
); );
#define CL_CHECK(err, name) \ #define CL_CHECK(err) \
do { \ do { \
cl_int err_ = (err); \ cl_int err_ = (err); \
if (err_ != CL_SUCCESS) { \ if (err_ != CL_SUCCESS) { \
fprintf(stderr, "ggml_opencl: %s error %d at %s:%d\n", name, err_, __FILE__, __LINE__); \ fprintf(stderr, "ggml_opencl: %s error %d at %s:%d\n", \
#err, err_, __FILE__, __LINE__); \
exit(1); \
} \
} while (0)
#define CLBLAST_CHECK(err) \
do { \
CLBlastStatusCode err_ = (err); \
if (err_ != CLBlastSuccess) { \
fprintf(stderr, "ggml_opencl: %s error %d at %s:%d\n", \
#err, err_, __FILE__, __LINE__); \
exit(1); \ exit(1); \
} \ } \
} while (0) } while (0)
@ -185,8 +196,7 @@ void ggml_cl_init(void) {
cl_platform_id platforms[NPLAT]; cl_platform_id platforms[NPLAT];
cl_uint num_platforms; cl_uint num_platforms;
err = clGetPlatformIDs(NPLAT, platforms, &num_platforms); CL_CHECK(clGetPlatformIDs(NPLAT, platforms, &num_platforms));
CL_CHECK(err, "clGetPlatformIDs");
char * GGML_OPENCL_PLATFORM = getenv("GGML_OPENCL_PLATFORM"); char * GGML_OPENCL_PLATFORM = getenv("GGML_OPENCL_PLATFORM");
if (GGML_OPENCL_PLATFORM != NULL) { if (GGML_OPENCL_PLATFORM != NULL) {
@ -197,11 +207,11 @@ void ggml_cl_init(void) {
exit(1); exit(1);
} else { } else {
platform = platforms[plat_num]; platform = platforms[plat_num];
clGetPlatformInfo(platform, CL_PLATFORM_NAME, sizeof(text_buffer), &text_buffer, NULL); CL_CHECK(clGetPlatformInfo(platform, CL_PLATFORM_NAME, sizeof(text_buffer), &text_buffer, NULL));
} }
} else { } else {
for (unsigned i = 0; i < num_platforms; i++) { for (unsigned i = 0; i < num_platforms; i++) {
clGetPlatformInfo(platforms[i], CL_PLATFORM_NAME, sizeof(text_buffer), &text_buffer, NULL); CL_CHECK(clGetPlatformInfo(platforms[i], CL_PLATFORM_NAME, sizeof(text_buffer), &text_buffer, NULL));
if (strstr(text_buffer, GGML_OPENCL_PLATFORM) != NULL) { if (strstr(text_buffer, GGML_OPENCL_PLATFORM) != NULL) {
platform = platforms[i]; platform = platforms[i];
break; break;
@ -220,8 +230,7 @@ void ggml_cl_init(void) {
if (GGML_OPENCL_DEVICE != NULL) { if (GGML_OPENCL_DEVICE != NULL) {
cl_device_id devices[NDEV]; cl_device_id devices[NDEV];
cl_uint num_devices; cl_uint num_devices;
clGetDeviceIDs(platform, CL_DEVICE_TYPE_ALL, NDEV, devices, &num_devices); CL_CHECK(clGetDeviceIDs(platform, CL_DEVICE_TYPE_ALL, NDEV, devices, &num_devices));
CL_CHECK(err, "clGetDeviceIDs");
unsigned dev_num; unsigned dev_num;
if (sscanf(GGML_OPENCL_DEVICE, " %u", &dev_num) == 1) { if (sscanf(GGML_OPENCL_DEVICE, " %u", &dev_num) == 1) {
@ -230,11 +239,11 @@ void ggml_cl_init(void) {
exit(1); exit(1);
} else { } else {
device = devices[dev_num]; device = devices[dev_num];
clGetDeviceInfo(device, CL_DEVICE_NAME, sizeof(text_buffer), &text_buffer, NULL); CL_CHECK(clGetDeviceInfo(device, CL_DEVICE_NAME, sizeof(text_buffer), &text_buffer, NULL));
} }
} else { } else {
for (unsigned i = 0; i < num_devices; i++) { for (unsigned i = 0; i < num_devices; i++) {
clGetDeviceInfo(devices[i], CL_DEVICE_NAME, sizeof(text_buffer), &text_buffer, NULL); CL_CHECK(clGetDeviceInfo(devices[i], CL_DEVICE_NAME, sizeof(text_buffer), &text_buffer, NULL));
if (strstr(text_buffer, GGML_OPENCL_DEVICE) != NULL) { if (strstr(text_buffer, GGML_OPENCL_DEVICE) != NULL) {
device = devices[i]; device = devices[i];
break; break;
@ -247,8 +256,7 @@ void ggml_cl_init(void) {
} else { } else {
fprintf(stderr, "ggml_opencl: selecting device: '%s'\n", text_buffer); fprintf(stderr, "ggml_opencl: selecting device: '%s'\n", text_buffer);
if (platform == NULL) { if (platform == NULL) {
err = clGetDeviceInfo(device, CL_DEVICE_PLATFORM, sizeof(&platform), &platform, NULL); CL_CHECK(clGetDeviceInfo(device, CL_DEVICE_PLATFORM, sizeof(&platform), &platform, NULL));
CL_CHECK(err, "clGetDeviceInfo");
} }
} }
} }
@ -258,8 +266,7 @@ void ggml_cl_init(void) {
cl_uint num_devices; cl_uint num_devices;
for (unsigned i = 0; i < num_platforms; i++) { for (unsigned i = 0; i < num_platforms; i++) {
clGetDeviceIDs(platforms[i], CL_DEVICE_TYPE_GPU, NDEV, devices, &num_devices); CL_CHECK(clGetDeviceIDs(platforms[i], CL_DEVICE_TYPE_GPU, NDEV, devices, &num_devices));
CL_CHECK(err, "clGetDeviceIDs");
if (num_devices > 0) { if (num_devices > 0) {
platform = platforms[i]; platform = platforms[i];
@ -278,24 +285,19 @@ void ggml_cl_init(void) {
}; };
if (device != NULL) { if (device != NULL) {
context = clCreateContext(properties, 1, &device, NULL, NULL, &err); CL_CHECK((context = clCreateContext(properties, 1, &device, NULL, NULL, &err), err));
CL_CHECK(err, "clCreateContext");
} else { } else {
context = clCreateContextFromType(properties, CL_DEVICE_TYPE_GPU, NULL, NULL, &err); CL_CHECK((context = clCreateContextFromType(properties, CL_DEVICE_TYPE_GPU, NULL, NULL, &err),
if (err == CL_DEVICE_NOT_AVAILABLE || err == CL_DEVICE_NOT_FOUND) { (err != CL_DEVICE_NOT_AVAILABLE && err != CL_DEVICE_NOT_FOUND ? err :
context = clCreateContextFromType(properties, CL_DEVICE_TYPE_DEFAULT, NULL, NULL, &err); (context = clCreateContextFromType(properties, CL_DEVICE_TYPE_DEFAULT, NULL, NULL, &err),
if (err == CL_DEVICE_NOT_AVAILABLE || err == CL_DEVICE_NOT_FOUND) { (err != CL_DEVICE_NOT_AVAILABLE && err != CL_DEVICE_NOT_FOUND ? err :
context = clCreateContextFromType(properties, CL_DEVICE_TYPE_ALL, NULL, NULL, &err); (context = clCreateContextFromType(properties, CL_DEVICE_TYPE_ALL, NULL, NULL, &err), err))
} ))
} ));
CL_CHECK(err, "clCreateContextFromType");
err = clGetContextInfo(context, CL_CONTEXT_DEVICES, sizeof(&device), &device, NULL);
CL_CHECK(err, "clGetContextInfo");
CL_CHECK(clGetContextInfo(context, CL_CONTEXT_DEVICES, sizeof(&device), &device, NULL));
if (platform == NULL) { if (platform == NULL) {
err = clGetDeviceInfo(device, CL_DEVICE_PLATFORM, sizeof(&platform), &platform, NULL); CL_CHECK(clGetDeviceInfo(device, CL_DEVICE_PLATFORM, sizeof(&platform), &platform, NULL));
CL_CHECK(err, "clGetDeviceInfo");
} }
} }
@ -308,26 +310,19 @@ void ggml_cl_init(void) {
clGetDeviceInfo(device, CL_DEVICE_NAME, sizeof(text_buffer), &text_buffer, NULL); clGetDeviceInfo(device, CL_DEVICE_NAME, sizeof(text_buffer), &text_buffer, NULL);
fprintf(stderr, "ggml_opencl: using device: '%s'\n", text_buffer); fprintf(stderr, "ggml_opencl: using device: '%s'\n", text_buffer);
CL_CHECK((queue = clCreateCommandQueue(context, device, CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE, &err),
queue = clCreateCommandQueue(context, device, CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE, &err); (err != CL_INVALID_PROPERTY && err != CL_INVALID_VALUE ? err :
if (err == CL_INVALID_PROPERTY || err == CL_INVALID_VALUE) { (queue = clCreateCommandQueue(context, device, 0, &err), err)
queue = clCreateCommandQueue(context, device, 0, &err); )));
}
CL_CHECK(err, "clCreateCommandQueue");
program = build_program_from_source(context, device, program_source); program = build_program_from_source(context, device, program_source);
// Prepare dequantize kernels // Prepare dequantize kernels
kernel_q4_0 = clCreateKernel(program, "dequantize_row_q4_0", &err); CL_CHECK((kernel_q4_0 = clCreateKernel(program, "dequantize_row_q4_0", &err), err));
CL_CHECK(err, "clCreateKernel"); CL_CHECK((kernel_q4_1 = clCreateKernel(program, "dequantize_row_q4_1", &err), err));
kernel_q4_1 = clCreateKernel(program, "dequantize_row_q4_1", &err); CL_CHECK((kernel_q5_0 = clCreateKernel(program, "dequantize_row_q5_0", &err), err));
CL_CHECK(err, "clCreateKernel"); CL_CHECK((kernel_q5_1 = clCreateKernel(program, "dequantize_row_q5_1", &err), err));
kernel_q5_0 = clCreateKernel(program, "dequantize_row_q5_0", &err); CL_CHECK((kernel_q8_0 = clCreateKernel(program, "dequantize_row_q8_0", &err), err));
CL_CHECK(err, "clCreateKernel");
kernel_q5_1 = clCreateKernel(program, "dequantize_row_q5_1", &err);
CL_CHECK(err, "clCreateKernel");
kernel_q8_0 = clCreateKernel(program, "dequantize_row_q8_0", &err);
CL_CHECK(err, "clCreateKernel");
} }
static void ggml_cl_malloc(size_t req_size, size_t* cur_size, cl_mem_flags flags, cl_mem* buf) { static void ggml_cl_malloc(size_t req_size, size_t* cur_size, cl_mem_flags flags, cl_mem* buf) {
@ -340,9 +335,8 @@ static void ggml_cl_malloc(size_t req_size, size_t* cur_size, cl_mem_flags flags
clReleaseMemObject(*buf); clReleaseMemObject(*buf);
} }
cl_int err; cl_int err;
*buf = clCreateBuffer(context, flags, req_size, NULL, &err); CL_CHECK((*buf = clCreateBuffer(context, flags, req_size, NULL, &err), err));
*cur_size = req_size; *cur_size = req_size;
CL_CHECK(err, "clCreateBuffer");
} }
void ggml_cl_sgemm_wrapper( void ggml_cl_sgemm_wrapper(
@ -351,7 +345,6 @@ void ggml_cl_sgemm_wrapper(
const float alpha, const void *host_a, const int lda, const float alpha, const void *host_a, const int lda,
const float *host_b, const int ldb, const float beta, const float *host_b, const int ldb, const float beta,
float *host_c, const int ldc, const int btype) { float *host_c, const int ldc, const int btype) {
cl_int err = 0;
cl_kernel kernel; cl_kernel kernel;
size_t global = n * k, local, size_qb; size_t global = n * k, local, size_qb;
@ -411,30 +404,26 @@ void ggml_cl_sgemm_wrapper(
cl_event ev_a, ev_qb, ev_b; cl_event ev_a, ev_qb, ev_b;
if (dequant) { if (dequant) {
err = clSetKernelArg(kernel, 0, sizeof(cl_mem), &cl_buffer_qb); CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), &cl_buffer_qb));
err |= clSetKernelArg(kernel, 1, sizeof(cl_mem), &cl_buffer_b); CL_CHECK(clSetKernelArg(kernel, 1, sizeof(cl_mem), &cl_buffer_b));
CL_CHECK(err, "clSetKernelArg"); CL_CHECK(clEnqueueWriteBuffer(queue, cl_buffer_qb, CL_FALSE, 0, size_qb, host_b, 0, NULL, &ev_qb));
err = clEnqueueWriteBuffer(queue, cl_buffer_qb, CL_FALSE, 0, size_qb, host_b, 0, NULL, &ev_qb);
CL_CHECK(err, "clEnqueueWriteBuffer qb");
} else { } else {
err = clEnqueueWriteBuffer(queue, cl_buffer_b, CL_FALSE, 0, size_b, host_b, 0, NULL, &ev_b); CL_CHECK(clEnqueueWriteBuffer(queue, cl_buffer_b, CL_FALSE, 0, size_b, host_b, 0, NULL, &ev_b));
CL_CHECK(err, "clEnqueueWriteBuffer b");
} }
err = clEnqueueWriteBuffer(queue, cl_buffer_a, CL_FALSE, 0, size_a, host_a, 0, NULL, &ev_a); CL_CHECK(clEnqueueWriteBuffer(queue, cl_buffer_a, CL_FALSE, 0, size_a, host_a, 0, NULL, &ev_a));
CL_CHECK(err, "clEnqueueWriteBuffer a");
if (dequant) { if (dequant) {
err = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &global, &local, 1, &ev_qb, &ev_b); CL_CHECK(clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &global, &local, 1, &ev_qb, &ev_b));
CL_CHECK(err, "clEnqueueNDRangeKernel"); CL_CHECK(clReleaseEvent(ev_qb));
clReleaseEvent(ev_qb);
} }
clWaitForEvents(1, &ev_a); CL_CHECK(clWaitForEvents(1, &ev_a));
clWaitForEvents(1, &ev_b); CL_CHECK(clWaitForEvents(1, &ev_b));
clReleaseEvent(ev_a); CL_CHECK(clReleaseEvent(ev_a));
clReleaseEvent(ev_b); CL_CHECK(clReleaseEvent(ev_b));
cl_event ev_sgemm; cl_event ev_sgemm;
CLBlastStatusCode status = CLBlastSgemm((CLBlastLayout)order, CLBLAST_CHECK(CLBlastSgemm(
(CLBlastLayout)order,
(CLBlastTranspose)trans_a, (CLBlastTranspose)trans_b, (CLBlastTranspose)trans_a, (CLBlastTranspose)trans_b,
m, n, k, m, n, k,
alpha, alpha,
@ -442,18 +431,13 @@ void ggml_cl_sgemm_wrapper(
cl_buffer_b, 0, ldb, cl_buffer_b, 0, ldb,
beta, beta,
cl_buffer_c, 0, ldc, cl_buffer_c, 0, ldc,
&queue, &ev_sgemm); &queue, &ev_sgemm));
if (status != CLBlastSuccess) {
fprintf(stderr, "Error: CLBlast SGEMM %d\n", status);
abort();
}
cl_event ev_c; cl_event ev_c;
clEnqueueReadBuffer(queue, cl_buffer_c, CL_TRUE, 0, size_c, host_c, 1, &ev_sgemm, &ev_c); CL_CHECK(clEnqueueReadBuffer(queue, cl_buffer_c, CL_TRUE, 0, size_c, host_c, 1, &ev_sgemm, &ev_c));
// Wait for completion // Wait for completion
clWaitForEvents(1, &ev_c); CL_CHECK(clWaitForEvents(1, &ev_c));
clReleaseEvent(ev_sgemm); CL_CHECK(clReleaseEvent(ev_sgemm));
clReleaseEvent(ev_c); CL_CHECK(clReleaseEvent(ev_c));
} }