vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

vulkan : implement Stable Diffusion operators (ggml/904)

Browse source 
* Fix Vulkan repeat op

* Implement Vulkan concat op

* Delete old Vulkan shader generator

* Implement Vulkan im2col op

* Implement Vulkan unary gelu_quick op

* Implement Vulkan group_norm op

* Implement Vulkan timestep_embedding op

* Implement Vulkan upscale op

* Fix Vulkan vk_context tensor extra index issue

* Fix Vulkan matmul shader parameter bug

* Properly fix Vulkan matmul shader parameter bug

* Add Vulkan ADD f16 + f32 -> f16 operator support

* Implement Vulkan tanh op

* Fix Vulkan group count too large Validation error on non-Nvidia GPUs

* Throw error when too much memory is requested

* Fix another Vulkan group count too large Validation error on non-Nvidia GPUs

* Fix matmul MMQ condition

* Implement Vulkan pad op

* Fix Vulkan crash when tensor is used multiple times in a compute graph

* Add Vulkan CONCAT f16 + f16 -> f16 op

* Add Vulkan LEAKY_RELU op
This commit is contained in:
0cc4m 2024-08-04 17:28:08 +02:00 • committed by Georgi Gerganov
parent 655858ace0
commit a3738b2fa7
28 changed files with 1032 additions and 293 deletions
Download patch file Download diff file Expand all files Collapse all files

35
ggml/src/vulkan-shaders/concat.comp Normal file
View file

@ -0,0 +1,35 @@
#version 450
#include "types.comp"
#include "generic_binary_head.comp"
void main() {
const uint idx = gl_GlobalInvocationID.z * 262144 + gl_GlobalInvocationID.y * 512 + gl_GlobalInvocationID.x;
const int dim = p.param3;
if (idx >= p.ne) {
return;
}
const uint i3 = idx / (p.ne22*p.ne21*p.ne20);
const uint i3_offset = i3 * p.ne22*p.ne21*p.ne20;
const uint i2 = (idx - i3_offset) / (p.ne21*p.ne20);
const uint i2_offset = i2*p.ne21*p.ne20;
const uint i1 = (idx - i3_offset - i2_offset) / p.ne20;
const uint i0 = idx - i3_offset - i2_offset - i1*p.ne20;
uint o[4] = {0, 0, 0, 0};
o[dim] = dim == 0 ? p.ne00 : (dim == 1 ? p.ne01 : (dim == 2 ? p.ne02 : p.ne03));
const uint src0_idx = i3*p.nb03 + i2*p.nb02 + i1*p.nb01 + i0*p.nb00;
const uint src1_idx = (i3 - o[3])*p.nb13 + (i2 - o[2])*p.nb12 + (i1 - o[1])*p.nb11 + (i0 - o[0])*p.nb10;
const uint dst_idx = i3*p.nb23 + i2*p.nb22 + i1*p.nb21 + i0*p.nb20;
const bool is_src0 = i0 < p.ne00 && i1 < p.ne01 && i2 < p.ne02 && i3 < p.ne03;
#ifndef OPTIMIZATION_ERROR_WORKAROUND
data_d[p.d_offset + dst_idx] = D_TYPE(is_src0 ? data_a[src0_idx] : data_b[src1_idx]);
#else
data_d[p.d_offset + dst_idx] = is_src0 ? data_a[src0_idx] : data_b[src1_idx];
#endif
}