ggml : build backends as libraries (#10256)
* ggml : build backends as libraries --------- Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> Co-authored-by: R0CKSTAR <xiaodong.ye@mthreads.com>
This commit is contained in:
parent
4a8ccb37ad
commit
ae8de6d50a
191 changed files with 17541 additions and 16879 deletions
60
ggml/src/ggml-vulkan/vulkan-shaders/generic_binary_head.comp
Normal file
60
ggml/src/ggml-vulkan/vulkan-shaders/generic_binary_head.comp
Normal file
|
@ -0,0 +1,60 @@
|
|||
#extension GL_EXT_shader_16bit_storage : require
|
||||
#extension GL_EXT_control_flow_attributes : require
|
||||
|
||||
layout (push_constant) uniform parameter
|
||||
{
|
||||
uint ne;
|
||||
uint ne00; uint ne01; uint ne02; uint ne03; uint nb00; uint nb01; uint nb02; uint nb03;
|
||||
uint ne10; uint ne11; uint ne12; uint ne13; uint nb10; uint nb11; uint nb12; uint nb13;
|
||||
uint ne20; uint ne21; uint ne22; uint ne23; uint nb20; uint nb21; uint nb22; uint nb23;
|
||||
uint d_offset;
|
||||
float param1; float param2; int param3;
|
||||
} p;
|
||||
|
||||
layout (binding = 0) readonly buffer A {A_TYPE data_a[];};
|
||||
layout (binding = 1) readonly buffer B {B_TYPE data_b[];};
|
||||
layout (binding = 2) writeonly buffer D {D_TYPE data_d[];};
|
||||
|
||||
// true if src0/src1 are the same shape and the indices can be reused without additional modulus
|
||||
layout(constant_id = 0) const bool norepeat = false;
|
||||
|
||||
uint get_idx() {
|
||||
return gl_GlobalInvocationID.z * 262144 + gl_GlobalInvocationID.y * 512 + gl_GlobalInvocationID.x;
|
||||
}
|
||||
|
||||
// mod and div are expensive and coordinates/dimensions are often power of 2 or equal to 1
|
||||
uint fastmod(uint a, uint b) {
|
||||
if ((b & (b-1)) == 0) {
|
||||
return a & (b-1);
|
||||
}
|
||||
return a % b;
|
||||
}
|
||||
|
||||
uint fastdiv(uint a, uint b) {
|
||||
return (a < b) ? 0 : (a / b);
|
||||
}
|
||||
|
||||
void get_indices(uint idx, out uint i00, out uint i01, out uint i02, out uint i03) {
|
||||
i03 = fastdiv(idx, (p.ne02*p.ne01*p.ne00));
|
||||
const uint i03_offset = i03 * p.ne02*p.ne01*p.ne00;
|
||||
i02 = fastdiv((idx - i03_offset), (p.ne01*p.ne00));
|
||||
const uint i02_offset = i02*p.ne01*p.ne00;
|
||||
i01 = (idx - i03_offset - i02_offset) / p.ne00;
|
||||
i00 = idx - i03_offset - i02_offset - i01*p.ne00;
|
||||
}
|
||||
|
||||
uint src0_idx(uint i00, uint i01, uint i02, uint i03) {
|
||||
return i03*p.nb03 + i02*p.nb02 + i01*p.nb01 + i00*p.nb00;
|
||||
}
|
||||
|
||||
uint src1_idx(uint i00, uint i01, uint i02, uint i03) {
|
||||
if (norepeat) {
|
||||
return i03*p.nb13 + i02*p.nb12 + i01*p.nb11 + i00*p.nb10;
|
||||
} else {
|
||||
return fastmod(i03, p.ne13)*p.nb13 + fastmod(i02, p.ne12)*p.nb12 + fastmod(i01, p.ne11)*p.nb11 + fastmod(i00, p.ne10)*p.nb10;
|
||||
}
|
||||
}
|
||||
|
||||
uint dst_idx(uint i00, uint i01, uint i02, uint i03) {
|
||||
return i03*p.nb23 + i02*p.nb22 + i01*p.nb21 + i00*p.nb20;
|
||||
}
|
Loading…
Add table
Add a link
Reference in a new issue