huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-16 18:48:51 +00:00
Code Issues Packages Projects Releases Wiki Activity

Also optimize the contiguous case for the binary cuda kernels.

Browse Source
This commit is contained in:
laurent
2023-06-23 19:04:13 +01:00
parent 5ca309ecb0
commit 88187b784b
1 changed files with 48 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
kernels/src/binary_op_macros.cuh
View File

@ -12,6 +12,41 @@ extern "C" __global__ void FN_NAME( \
const size_t *dims = dims_and_strides; \
const size_t *lhs_strides = dims_and_strides + 1 * num_dims; \
const size_t *rhs_strides = dims_and_strides + 2 * num_dims; \
bool lhs_cont = is_contiguous(num_dims, dims, lhs_strides); \
bool rhs_cont = is_contiguous(num_dims, dims, rhs_strides); \
if (lhs_cont && rhs_cont) { \
for (unsigned int i = blockIdx.x * blockDim.x + threadIdx.x; i < numel; i += blockDim.x * gridDim.x) { \
TYPENAME x = lhs ? lhs[i] : out[i]; \
TYPENAME y = rhs ? rhs[i] : out[i]; \
out[i] = FUNC; \
} \
} else if (lhs_cont) { \
for (unsigned int i = blockIdx.x * blockDim.x + threadIdx.x; i < numel; i += blockDim.x * gridDim.x) { \
unsigned int tmp_i = i; \
unsigned int rhs_i = 0; \
for (int d = num_dims - 1; d >= 0; d--) { \
unsigned int i_dim = tmp_i % dims[d]; \
rhs_i += i_dim * rhs_strides[d]; \
tmp_i /= dims[d]; \
} \
TYPENAME x = lhs ? lhs[i] : out[i]; \
TYPENAME y = rhs ? rhs[rhs_i] : out[i]; \
out[i] = FUNC; \
} \
} else if (rhs_cont) { \
for (unsigned int i = blockIdx.x * blockDim.x + threadIdx.x; i < numel; i += blockDim.x * gridDim.x) { \
unsigned int tmp_i = i; \
unsigned int lhs_i = 0; \
for (int d = num_dims - 1; d >= 0; d--) { \
unsigned int i_dim = tmp_i % dims[d]; \
lhs_i += i_dim * lhs_strides[d]; \
tmp_i /= dims[d]; \
} \
TYPENAME x = lhs ? lhs[lhs_i] : out[i]; \
TYPENAME y = rhs ? rhs[i] : out[i]; \
out[i] = FUNC; \
} \
} else { \
for (unsigned int i = blockIdx.x * blockDim.x + threadIdx.x; i < numel; i += blockDim.x * gridDim.x) { \
unsigned int tmp_i = i; \
unsigned int lhs_i = 0; \
@ -26,4 +61,5 @@ extern "C" __global__ void FN_NAME( \
TYPENAME y = rhs ? rhs[rhs_i] : out[i]; \
out[i] = FUNC; \
} \
} \
} \