Skip to content

[Core] Thread-safe code generation for BitNet kernels #386

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

Open
DT-Foss wants to merge 1 commit into
base: main
Choose a base branch
from
+23 −8
Open

[Core] Thread-safe code generation for BitNet kernels #386

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
15 changes: 11 additions & 4 deletions utils/codegen_tl1.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,10 +5,11 @@
def gen_ctor_code():
kernel_code = "\n\
#include \"ggml-bitnet.h\"\n\
#include <atomic>\n\
#define GGML_BITNET_MAX_NODES 8192\n\
static bool initialized = false;\n\
static bitnet_tensor_extra * bitnet_tensor_extras = nullptr;\n\
static size_t bitnet_tensor_extras_index = 0;\n\
static std::atomic<size_t> bitnet_tensor_extras_index{0};\n\
static void * aligned_malloc(size_t size) {{\n\
#if defined(_WIN32)\n\
return _aligned_malloc(size, 64);\n\
Expand Down Expand Up @@ -355,14 +356,20 @@ def gen_transform_code(kernel_shape):
float * i2_scales = (float * )(qweights + k * m / 4);\n\
scales[0] = (bitnet_float_type) i2_scales[0];\n\
\n\
tensor->extra = bitnet_tensor_extras + bitnet_tensor_extras_index;\n\
bitnet_tensor_extras[bitnet_tensor_extras_index++] = {\n\
size_t current_index = bitnet_tensor_extras_index.fetch_add(1);\n\
if (current_index >= GGML_BITNET_MAX_NODES) {{\n\
fprintf(stderr, \"ggml_bitnet_transform_tensor: GGML_BITNET_MAX_NODES reached (%d)\\n\", GGML_BITNET_MAX_NODES);\n\
return;\n\
}}\n\
\n\
tensor->extra = bitnet_tensor_extras + current_index;\n\
bitnet_tensor_extras[current_index] = {{\n\
/* .lut_scales_size = */ lut_scales_size,\n\
/* .BK = */ BK,\n\
/* .n_tile_num = */ n_tile_num,\n\
/* .qweights = */ qweights,\n\
/* .scales = */ scales\n\
};\n\
}};\n\
}\n"])

return kernel_code
Expand Down
16 changes: 12 additions & 4 deletions utils/codegen_tl2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,11 +6,13 @@ def gen_ctor_code():
kernel_code = "\n\
#include \"ggml-bitnet.h\"\n\
#include <cstring>\n\
#include <cstdio>\n\
#include <atomic>\n\
#include <immintrin.h>\n\
#define GGML_BITNET_MAX_NODES 8192\n\
static bool initialized = false;\n\
static bitnet_tensor_extra * bitnet_tensor_extras = nullptr;\n\
static size_t bitnet_tensor_extras_index = 0;\n\
static std::atomic<size_t> bitnet_tensor_extras_index{{0}};\n\
static void * aligned_malloc(size_t size) {\n\
#if defined(_WIN32)\n\
return _aligned_malloc(size, 64);\n\
Expand Down Expand Up @@ -661,14 +663,20 @@ def gen_transform_code(kernel_shapes):
float * i2_scales = (float * )(qweights + nbytes);\n\
scales[0] = (bitnet_float_type) i2_scales[0];\n\
\n\
tensor->extra = bitnet_tensor_extras + bitnet_tensor_extras_index;\n\
bitnet_tensor_extras[bitnet_tensor_extras_index++] = {\n\
size_t current_index = bitnet_tensor_extras_index.fetch_add(1);\n\
if (current_index >= GGML_BITNET_MAX_NODES) {{\n\
fprintf(stderr, \"ggml_bitnet_transform_tensor: GGML_BITNET_MAX_NODES reached (%d)\\n\", GGML_BITNET_MAX_NODES);\n\
return;\n\
}}\n\
\n\
tensor->extra = bitnet_tensor_extras + current_index;\n\
bitnet_tensor_extras[current_index] = {{\n\
/* .lut_scales_size = */ lut_scales_size,\n\
/* .BK = */ BK,\n\
/* .n_tile_num = */ n_tile_num,\n\
/* .qweights = */ qweights,\n\
/* .scales = */ scales\n\
};\n\
}};\n\
}\n"])

return kernel_code
Expand Down