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

remove a multiply

Browse source 
polynomial iq4_nl test (slower but keep as reference)
This commit is contained in:
Eve 2024-12-04 16:31:14 -05:00
parent b7ad234517
commit 5fbaf121db
2 changed files with 9 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

7
ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs.comp
View file

@ -100,6 +100,11 @@ vec4 dequantize4(uint ib, uint iqs, uint a_offset) {
#endif
#if defined(DATA_A_IQ4_NL)
float iq_helper(uint i) {
const float x = float(i);
return round(((0.080958*x-1.875836)*x+25.907107)*x-127.663571);
}
vec2 dequantize(uint ib, uint iqs, uint a_offset) {
const float d = float(data_a[a_offset + ib].d);
const uint vui = uint(data_a[a_offset + ib].qs[iqs]);
@ -107,6 +112,6 @@ vec2 dequantize(uint ib, uint iqs, uint a_offset) {
}
vec4 dequantize4(uint ib, uint iqs, uint a_offset) {
const uint vui = uint(data_a_packed16[a_offset + ib].qs[iqs/2]);
return vec4(kvalues_iq4nl[vui & 0xF], kvalues_iq4nl[(vui >> 4) & 0xF], kvalues_iq4nl[(vui >> 8) & 0xF], kvalues_iq4nl[(vui >> 12) & 0xF]);
return vec4(iq_helper(vui & 0xF), iq_helper((vui >> 4) & 0xF), iq_helper((vui >> 8) & 0xF), iq_helper((vui >> 12) & 0xF));
}
#endif

4
ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec.comp
View file

@ -53,8 +53,10 @@ void iter(inout FLOAT_TYPE temp[NUM_ROWS], const uint first_row, const uint num_
b1 = FLOAT_TYPE(data_b[b_offset + iybs + iqs + y_offset]);
}
#endif
uint ibi = first_row*p.ncols;
[[unroll]] for (uint n = 0; n < num_rows; ++n) {
const uint ib = ((first_row + n)*p.ncols + col)/QUANT_K; // block index
const uint ib = (ibi + col)/QUANT_K; // block index
ibi += p.ncols;
#if K_PER_ITER == 8
// TODO: can we dequant as f16 instead of as vec?