Skip to content

Rearranging calculations in quantized linear tiled implementaiton to reduce register usage and gain performance. #15111

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Oct 16, 2025
+31 −33
Merged

Rearranging calculations in quantized linear tiled implementaiton to reduce register usage and gain performance. #15111

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
64 changes: 31 additions & 33 deletions backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw_tiled.glsl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -66,8 +66,6 @@ void main() {
return;
}

T mat1[TILE_ROWS][4];
VEC4_T qmat2[4][TILE_TXCOLS];
VEC4_T sums[TILE_ROWS][TILE_TXCOLS];

VEC4_T scales[TILE_TXCOLS];
Expand All @@ -86,12 +84,35 @@ void main() {
for (uint16_t pos = uint16_t(0), txpos = uint16_t(0);
pos < uint16_t(in_sizes.x);
pos += uint16_t(4), txpos += uint16_t(1)) {

T mat1[TILE_ROWS][4];

$if IN_STORAGE == "buffer":
uint in_row_txstride = div4(in_sizes.x);

// Preload input tensor
for (int i = 0; i < TILE_ROWS; i++) {
$if IN_STORAGE == "buffer":
VEC4_T tmp = t_in[(out_row + i) * in_row_txstride + txpos];
mat1[i][0] = tmp.x;
mat1[i][1] = tmp.y;
mat1[i][2] = tmp.z;
mat1[i][3] = tmp.w;
$else:
VEC4_T tmp = VEC4_T(texelFetch(t_in, u16vec3(txpos, out_row + i, 0), 0));
mat1[i][0] = tmp.x;
mat1[i][1] = tmp.y;
mat1[i][2] = tmp.z;
mat1[i][3] = tmp.w;
}

$if WEIGHT_STORAGE == "buffer":
uint qmat2_bufi;
uint weight_row_txstride = div4(weight_sizes.x);

// Preload weight tensor
for (int r = 0; r < 4; r++) {
VEC4_T qmat2[TILE_TXCOLS];
$if QUANT_NBITS == 4:
$for c in range(0, TILE_TXCOLS, 2):
$if WEIGHT_STORAGE == "buffer":
Expand All @@ -101,44 +122,21 @@ void main() {
const uvec4 packed_weight_tex = texelFetch(
t_weight, u16vec2(weight_txcol + ${c}, pos + r), 0);

qmat2[r][${c}] = (VEC4_T((packed_weight_tex & 0xF0) >> 4) - 8.0);
qmat2[r][${c + 1}] = (VEC4_T(packed_weight_tex & 0x0F) - 8.0);
qmat2[${c}] = (VEC4_T((packed_weight_tex & 0xF0) >> 4) - 8.0);
qmat2[${c + 1}] = (VEC4_T(packed_weight_tex & 0x0F) - 8.0);
$else:
$for c in range(TILE_TXCOLS):
$if WEIGHT_STORAGE == "buffer":
qmat2_bufi = (pos + r) * weight_row_txstride + out_txcol;
qmat2[r][${c}] = t_weight[qmat2_bufi + ${c}];
qmat2[${c}] = t_weight[qmat2_bufi + ${c}];
$else:
qmat2[r][${c}] = VEC4_T(
qmat2[${c}] = VEC4_T(
texelFetch(t_weight, u16vec2(out_txcol + ${c}, pos + r), 0));
}

$if IN_STORAGE == "buffer":
uint in_row_txstride = div4(in_sizes.x);

// Preload input tensor
for (int i = 0; i < TILE_ROWS; i++) {
$if IN_STORAGE == "buffer":
VEC4_T tmp = t_in[(out_row + i) * in_row_txstride + txpos];
mat1[i][0] = tmp.x;
mat1[i][1] = tmp.y;
mat1[i][2] = tmp.z;
mat1[i][3] = tmp.w;
$else:
VEC4_T tmp = VEC4_T(texelFetch(t_in, u16vec3(txpos, out_row + i, 0), 0));
mat1[i][0] = tmp.x;
mat1[i][1] = tmp.y;
mat1[i][2] = tmp.z;
mat1[i][3] = tmp.w;
}

// Accumulate output
for (int r = 0; r < TILE_ROWS; ++r) {
$for c in range(TILE_TXCOLS):
sums[r][${c}] += mat1[r][0] * qmat2[0][${c}] +
mat1[r][1] * qmat2[1][${c}] +
mat1[r][2] * qmat2[2][${c}] +
mat1[r][3] * qmat2[3][${c}];
for (int tr = 0; tr < TILE_ROWS; ++tr) {
$for c in range(TILE_TXCOLS):
sums[tr][${c}] += qmat2[${c}] * mat1[tr][r];
}
}
}

Expand Down
Loading