Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

vulkan: support non-contig unary/glu ops (#24215)

Browse Source 
* vulkan: support non-contig unary/glu ops

Change unary/glu ops to pass in all strides and use fastdiv for the index
calculation. Put all unary ops in one file, similar to glu, to share the
code. codex went ahead and added expm1 without me asking, but I had to
make it do a real precision analysis rather than just making stuff up.

unary.comp initially couldn't use generic_unary_head because there wasn't
space for xielu's additional constants. Fixing this required packing the
fastdiv 'L' values.

* attempt to workaround compiler bug

* resolve conflict from #23991

* use expm1
This commit is contained in:
Jeff Bolz
2026-06-13 08:44:15 -05:00
committed by GitHub
parent 597b6672e8
commit 1a7718b4c5
33 changed files with 7553 additions and 4876 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-vulkan/vulkan-shaders/generic_unary_head.glsl
+21 -19
View File
@@ -7,14 +7,12 @@ layout (push_constant) uniform parameter
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 misalign_offsets;
float param1; float param2;
float param1; float param2; float param3; float param4;
uint ne0_012mp; uint ne0_012L;
uint ne0_01mp; uint ne0_01L;
uint ne0_0mp; uint ne0_0L;
uint ne1_012mp; uint ne1_012L;
uint ne1_01mp; uint ne1_01L;
uint ne1_0mp; uint ne1_0L;
// The three L values are packed as bytes to keep this layout under the 128B
// push constant limit while still leaving room for four float parameters.
uint ne0_012mp; uint ne0_01mp; uint ne0_0mp; uint ne0_Ls;
uint ne1_012mp; uint ne1_01mp; uint ne1_0mp; uint ne1_Ls;
} p;
layout (binding = 0) readonly buffer A {A_TYPE data_a[];};
@@ -42,42 +40,46 @@ uint fastdiv(uint n, uint mp, uint L) {
return (msbs + n) >> L;
}
uint fastdiv_L(uint packed, uint slot) {
return (packed >> (slot * 8)) & 0x3Fu;
}
uint src0_idx(uint idx) {
const uint i03 = fastdiv(idx, p.ne0_012mp, p.ne0_012L);
const uint i03 = fastdiv(idx, p.ne0_012mp, fastdiv_L(p.ne0_Ls, 0));
const uint i03_offset = i03 * p.ne02*p.ne01*p.ne00;
const uint i02 = fastdiv(idx - i03_offset, p.ne0_01mp, p.ne0_01L);
const uint i02 = fastdiv(idx - i03_offset, p.ne0_01mp, fastdiv_L(p.ne0_Ls, 1));
const uint i02_offset = i02*p.ne01*p.ne00;
const uint i01 = fastdiv(idx - i03_offset - i02_offset, p.ne0_0mp, p.ne0_0L);
const uint i01 = fastdiv(idx - i03_offset - i02_offset, p.ne0_0mp, fastdiv_L(p.ne0_Ls, 2));
const uint i00 = idx - i03_offset - i02_offset - i01*p.ne00;
return i03*p.nb03 + i02*p.nb02 + i01*p.nb01 + i00*p.nb00;
}
uint dst_idx(uint idx) {
const uint i13 = fastdiv(idx, p.ne1_012mp, p.ne1_012L);
const uint i13 = fastdiv(idx, p.ne1_012mp, fastdiv_L(p.ne1_Ls, 0));
const uint i13_offset = i13 * p.ne12*p.ne11*p.ne10;
const uint i12 = fastdiv(idx - i13_offset, p.ne1_01mp, p.ne1_01L);
const uint i12 = fastdiv(idx - i13_offset, p.ne1_01mp, fastdiv_L(p.ne1_Ls, 1));
const uint i12_offset = i12*p.ne11*p.ne10;
const uint i11 = fastdiv(idx - i13_offset - i12_offset, p.ne1_0mp, p.ne1_0L);
const uint i11 = fastdiv(idx - i13_offset - i12_offset, p.ne1_0mp, fastdiv_L(p.ne1_Ls, 2));
const uint i10 = idx - i13_offset - i12_offset - i11*p.ne10;
return i13*p.nb13 + i12*p.nb12 + i11*p.nb11 + i10*p.nb10;
}
uint src0_idx_quant(uint idx, uint qk) {
const uint i03 = fastdiv(idx, p.ne0_012mp, p.ne0_012L);
const uint i03 = fastdiv(idx, p.ne0_012mp, fastdiv_L(p.ne0_Ls, 0));
const uint i03_offset = i03 * p.ne02*p.ne01*p.ne00;
const uint i02 = fastdiv(idx - i03_offset, p.ne0_01mp, p.ne0_01L);
const uint i02 = fastdiv(idx - i03_offset, p.ne0_01mp, fastdiv_L(p.ne0_Ls, 1));
const uint i02_offset = i02*p.ne01*p.ne00;
const uint i01 = fastdiv(idx - i03_offset - i02_offset, p.ne0_0mp, p.ne0_0L);
const uint i01 = fastdiv(idx - i03_offset - i02_offset, p.ne0_0mp, fastdiv_L(p.ne0_Ls, 2));
const uint i00 = idx - i03_offset - i02_offset - i01*p.ne00;
return i03*p.nb03 + i02*p.nb02 + i01*p.nb01 + (i00/qk)*p.nb00;
}
uint dst_idx_quant(uint idx, uint qk) {
const uint i13 = fastdiv(idx, p.ne1_012mp, p.ne1_012L);
const uint i13 = fastdiv(idx, p.ne1_012mp, fastdiv_L(p.ne1_Ls, 0));
const uint i13_offset = i13 * p.ne12*p.ne11*p.ne10;
const uint i12 = fastdiv(idx - i13_offset, p.ne1_01mp, p.ne1_01L);
const uint i12 = fastdiv(idx - i13_offset, p.ne1_01mp, fastdiv_L(p.ne1_Ls, 1));
const uint i12_offset = i12*p.ne11*p.ne10;
const uint i11 = fastdiv(idx - i13_offset - i12_offset, p.ne1_0mp, p.ne1_0L);
const uint i11 = fastdiv(idx - i13_offset - i12_offset, p.ne1_0mp, fastdiv_L(p.ne1_Ls, 2));
const uint i10 = idx - i13_offset - i12_offset - i11*p.ne10;
return i13*p.nb13 + i12*p.nb12 + i11*p.nb11 + (i10/qk)*p.nb10;
}