vulkan: Implement GGML_OP_CUMSUM (#17479)

This commit is contained in:
Jeff Bolz
2025-11-26 00:08:10 -06:00
committed by GitHub
parent 583cb83416
commit b3b03a7baf
5 changed files with 125 additions and 24 deletions
@@ -1,6 +1,7 @@
#version 450
#include "types.glsl"
#include "sum_rows.glsl"
#extension GL_EXT_control_flow_attributes : enable
@@ -11,30 +12,6 @@ layout (binding = 1) writeonly buffer D {D_TYPE data_d[];};
layout (constant_id = 0) const uint BLOCK_SIZE = 32;
layout (push_constant) uniform parameter
{
uint n_cols;
uint ne01, ne02;
uint nb01, nb02, nb03;
uint nb11, nb12, nb13;
float weight;
uint misalign_offsets;
uint ne0_12mp, ne0_12L;
uint ne0_1mp, ne0_1L;
} p;
uint get_aoffset() { return p.misalign_offsets >> 16; }
uint get_doffset() { return p.misalign_offsets & 0xFFFF; }
// see init_fastdiv_values in ggml-vulkan.cpp
uint fastdiv(uint n, uint mp, uint L) {
uint msbs, lsbs;
// msbs = mulhi(n, mp)
umulExtended(n, mp, msbs, lsbs);
return (msbs + n) >> L;
}
shared FLOAT_TYPE tmp[BLOCK_SIZE];
void main() {