diff options
Diffstat (limited to 'ggml/src/vulkan-shaders/mul_mat_vec.comp')
| -rw-r--r-- | ggml/src/vulkan-shaders/mul_mat_vec.comp | 13 |
1 files changed, 10 insertions, 3 deletions
diff --git a/ggml/src/vulkan-shaders/mul_mat_vec.comp b/ggml/src/vulkan-shaders/mul_mat_vec.comp index 15d2a806..46a6369b 100644 --- a/ggml/src/vulkan-shaders/mul_mat_vec.comp +++ b/ggml/src/vulkan-shaders/mul_mat_vec.comp @@ -16,6 +16,13 @@ void main() { const uint row = gl_WorkGroupID.x + gl_NumWorkGroups.x * gl_WorkGroupID.z; const uint tid = gl_LocalInvocationID.x; + // There are not enough cols to use all threads + if (tid >= p.ncols) { + return; + } + + const uint block_size = min(p.ncols, BLOCK_SIZE); + uint a_offset, b_offset, d_offset; get_offsets(a_offset, b_offset, d_offset); @@ -23,8 +30,8 @@ void main() { tmp[tid] = FLOAT_TYPE(0.0f); - [[unroll]] for (uint i = 0; i < p.ncols/BLOCK_SIZE; i += 2) { - const uint col = i*BLOCK_SIZE + 2*tid; + [[unroll]] for (uint i = 0; i < p.ncols/block_size; i += 2) { + const uint col = i*block_size + 2*tid; const uint ib = (row*p.ncols + col)/QUANT_K; // block index const uint iqs = (col%QUANT_K)/QUANT_R; // quant index const uint iybs = col - col%QUANT_K; // y block start index @@ -38,7 +45,7 @@ void main() { // sum up partial sums and write back result barrier(); - [[unroll]] for (uint s = BLOCK_SIZE/2; s > 0; s >>= 1) { + [[unroll]] for (uint s = block_size/2; s > 0; s >>= 1) { if (tid < s) { tmp[tid] += tmp[tid + s]; } |