huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-19 11:56:45 +00:00
Code Issues Packages Projects Releases Wiki Activity

Avoid using tmp values. (#609)

Browse Source
This commit is contained in:
Laurent Mazare
2023-08-26 12:28:28 +01:00
committed by GitHub
parent c72eb3d75b
commit 06b37ea7ad
1 changed files with 8 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
candle-core/src/quantized/k_quants.rs
View File

@ -916,8 +916,8 @@ impl GgmlType for BlockQ4K {
const KMASK3: u32 = 0x03030303; const KMASK3: u32 = 0x03030303;
let mut utmp: [u32; 4] = [0; 4]; let mut utmp: [u32; 4] = [0; 4];
let mut scales: [u8; 8]; let mut scales: [u8; 8] = [0; 8];
let mut mins: [u8; 8]; let mut mins: [u8; 8] = [0; 8];
let mut aux8: [i8; QK_K] = [0; QK_K]; let mut aux8: [i8; QK_K] = [0; QK_K];
let mut aux16: [i16; 8] = [0; 8]; let mut aux16: [i16; 8] = [0; 8];
@ -953,14 +953,8 @@ impl GgmlType for BlockQ4K {
utmp[0] &= KMASK1; utmp[0] &= KMASK1;
//extract scales and mins //extract scales and mins
let mut utemp_scales = &mut [0u32; 2]; LittleEndian::write_u32_into(&utmp[0..2], &mut scales);
let mut utemp_mins = &mut [0u32; 2]; LittleEndian::write_u32_into(&utmp[2..4], &mut mins);
utemp_scales.copy_from_slice(&utmp[0..2]);
utemp_mins.copy_from_slice(&utmp[2..4]);
scales =
unsafe { *std::mem::transmute::<&mut [u32; 2], &mut [u8; 8]>(&mut utemp_scales) };
mins = unsafe { *std::mem::transmute::<&mut [u32; 2], &mut [u8; 8]>(&mut utemp_mins) };
let mut sumi = 0; let mut sumi = 0;
for j in 0..QK_K / 16 { for j in 0..QK_K / 16 {
@ -1101,8 +1095,8 @@ impl GgmlType for BlockQ5K {
const KMASK3: u32 = 0x03030303; const KMASK3: u32 = 0x03030303;
let mut utmp: [u32; 4] = [0; 4]; let mut utmp: [u32; 4] = [0; 4];
let mut scales: [u8; 8]; let mut scales: [u8; 8] = [0; 8];
let mut mins: [u8; 8]; let mut mins: [u8; 8] = [0; 8];
let mut aux8: [i8; QK_K] = [0; QK_K]; let mut aux8: [i8; QK_K] = [0; QK_K];
let mut aux16: [i16; 8] = [0; 8]; let mut aux16: [i16; 8] = [0; 8];
@ -1145,14 +1139,8 @@ impl GgmlType for BlockQ5K {
utmp[0] &= KMASK1; utmp[0] &= KMASK1;
//extract scales and mins //extract scales and mins
let mut utemp_scales = &mut [0u32; 2]; LittleEndian::write_u32_into(&utmp[0..2], &mut scales);
let mut utemp_mins = &mut [0u32; 2]; LittleEndian::write_u32_into(&utmp[2..4], &mut mins);
utemp_scales.copy_from_slice(&utmp[0..2]);
utemp_mins.copy_from_slice(&utmp[2..4]);
scales =
unsafe { *std::mem::transmute::<&mut [u32; 2], &mut [u8; 8]>(&mut utemp_scales) };
mins = unsafe { *std::mem::transmute::<&mut [u32; 2], &mut [u8; 8]>(&mut utemp_mins) };
let mut sumi = 0; let mut sumi = 0;
for j in 0..QK_K / 16 { for j in 0..QK_K / 16 {