Skip to content

vulkan: clamp matmul and FA results to the max finite value #15652

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 2 commits into from
Aug 31, 2025
Merged

vulkan: clamp matmul and FA results to the max finite value #15652

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
4387882
vulkan: clamp matmul and FA results to the max finite value
jeffbolznv Aug 29, 2025
b87d4ef
only clamp for fp16
jeffbolznv Aug 30, 2025
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
1 change: 1 addition & 0 deletions ggml/src/ggml-vulkan/vulkan-shaders/flash_attn.comp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -334,6 +334,7 @@ void main() {
[[unroll]] for (uint32_t d = 0; d < HSV_per_thread / 4; ++d) {
[[unroll]] for (uint32_t r = 0; r < Br; ++r) {
Of[r][d] *= Lfrcp[r];
Of[r][d] = clamp(Of[r][d], -vec4(ACC_TYPE_MAX), vec4(ACC_TYPE_MAX));
}
}

Expand Down
1 change: 1 addition & 0 deletions ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm1.comp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -373,6 +373,7 @@ void main() {
[[unroll]] for (uint32_t d = 0; d < HSV_per_thread / 4; ++d) {
[[unroll]] for (uint32_t r = 0; r < rows_per_thread; ++r) {
Of[r][d] *= ACC_TYPE(Lfrcp[r]);
Of[r][d] = clamp(Of[r][d], -ACC_TYPE_MAX, ACC_TYPE_MAX);
}
}

Expand Down
2 changes: 2 additions & 0 deletions ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -283,6 +283,8 @@ void main() {

O = Ldiag*O;

[[unroll]] for (uint i = 0; i < O.length(); ++i) { O[i] = clamp(O[i], -ACC_TYPE_MAX, ACC_TYPE_MAX); }

uint32_t o_offset = iq3*p.ne2*p.ne1*HSV;

coopmat<D_TYPE, gl_ScopeWorkgroup, Br, HSV_pad, gl_MatrixUseAccumulator> O_D = coopmat<D_TYPE, gl_ScopeWorkgroup, Br, HSV_pad, gl_MatrixUseAccumulator>(O);
Expand Down
4 changes: 4 additions & 0 deletions ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_split_k_reduce.comp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -111,6 +111,10 @@ void main() {
}
}
O *= L;

const float FLT_MAX = uintBitsToFloat(0x7F7FFFFF);
O = clamp(O, -FLT_MAX, FLT_MAX);

data_d[iq3 * D * N + D * n + d] = O;
}
}
12 changes: 12 additions & 0 deletions ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -891,6 +891,18 @@ void main() {
barrier();
}

#ifdef COOPMAT
[[unroll]] for (uint j = 0; j < cms_per_row * cms_per_col; j++) {
[[unroll]] for (uint i = 0; i < sums[j].length(); ++i) {
sums[j][i] = clamp(sums[j][i], -ACC_TYPE_MAX, ACC_TYPE_MAX);
}
}
#else
[[unroll]] for (uint i = 0; i < WMITER*TM*WNITER*TN; i++) {
sums[i] = clamp(sums[i], -ACC_TYPE_MAX, ACC_TYPE_MAX);
}
#endif

const uint dr = ir * BM + warp_r * WM;
const uint dc = ic * BN + warp_c * WN;

Expand Down
7 changes: 7 additions & 0 deletions ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -349,6 +349,8 @@ void main() {
sum = coopMatMulAdd(mat_a, mat_b, sum);
block_k += BK;
}
[[unroll]] for (uint i = 0; i < sum.length(); ++i) { sum[i] = clamp(sum[i], -ACC_TYPE_MAX, ACC_TYPE_MAX); }

coopmat<D_TYPE, gl_ScopeWorkgroup, BM, BNover4, gl_MatrixUseAccumulator> mat_d = coopmat<D_TYPE, gl_ScopeWorkgroup, BM, BNover4, gl_MatrixUseAccumulator>(sum);

coopMatStoreTensorNV(mat_d, data_d, pos_d, sliceTensorLayoutNV(tensorLayoutD, ic * BN, BNover4, ir * BM, BM), tensorViewTranspose);
Expand Down Expand Up @@ -388,6 +390,8 @@ void main() {
sum = coopMatMulAdd(mat_a, mat_b, sum);
block_k += BK;
}
[[unroll]] for (uint i = 0; i < sum.length(); ++i) { sum[i] = clamp(sum[i], -ACC_TYPE_MAX, ACC_TYPE_MAX); }

coopmat<D_TYPE, gl_ScopeWorkgroup, BM, BNover2, gl_MatrixUseAccumulator> mat_d = coopmat<D_TYPE, gl_ScopeWorkgroup, BM, BNover2, gl_MatrixUseAccumulator>(sum);

coopMatStoreTensorNV(mat_d, data_d, pos_d, sliceTensorLayoutNV(tensorLayoutD, ic * BN, BNover2, ir * BM, BM), tensorViewTranspose);
Expand Down Expand Up @@ -428,6 +432,8 @@ void main() {
sum = coopMatMulAdd(mat_a, mat_b, sum);
block_k += BK;
}
[[unroll]] for (uint i = 0; i < sum.length(); ++i) { sum[i] = clamp(sum[i], -ACC_TYPE_MAX, ACC_TYPE_MAX); }

coopmat<D_TYPE, gl_ScopeWorkgroup, BM, BN, gl_MatrixUseAccumulator> mat_d = coopmat<D_TYPE, gl_ScopeWorkgroup, BM, BN, gl_MatrixUseAccumulator>(sum);

coopMatStoreTensorNV(mat_d, data_d, pos_d, sliceTensorLayoutNV(tensorLayoutD, ic * BN, BN, ir * BM, BM), tensorViewTranspose);
Expand Down Expand Up @@ -485,6 +491,7 @@ void main() {
sum = coopMatMulAdd(mat_a, mat_b, sum);
}
}
[[unroll]] for (uint i = 0; i < sum.length(); ++i) { sum[i] = clamp(sum[i], -ACC_TYPE_MAX, ACC_TYPE_MAX); }

// Convert from ACC_TYPE to D_TYPE
coopmat<D_TYPE, gl_ScopeWorkgroup, BM, BN, gl_MatrixUseAccumulator> mat_d;
Expand Down
19 changes: 11 additions & 8 deletions ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -323,6 +323,7 @@ void matmul_shaders(bool fp16, MatMulIdType matmul_id_type, bool coopmat, bool c
}

base_dict["ACC_TYPE"] = f16acc ? "float16_t" : "float";
base_dict["ACC_TYPE_MAX"] = f16acc ? "\"float16_t(65504.0)\"" : "\"uintBitsToFloat(0x7F7FFFFF)\"";

if (coopmat) {
base_dict["COOPMAT"] = "1";
Expand Down Expand Up @@ -437,8 +438,10 @@ void process_shaders() {

// flash attention
for (const auto& f16acc : {false, true}) {
std::string acctype = f16acc ? "float16_t" : "float";
std::string acctypev4 = f16acc ? "f16vec4" : "vec4";
std::map<std::string, std::string> fa_base_dict = base_dict;
fa_base_dict["ACC_TYPE"] = f16acc ? "float16_t" : "float";
fa_base_dict["ACC_TYPEV4"] = f16acc ? "f16vec4" : "vec4";
fa_base_dict["ACC_TYPE_MAX"] = f16acc ? "\"float16_t(65504.0)\"" : "\"uintBitsToFloat(0x7F7FFFFF)\"";

for (const auto& tname : type_names) {
if (tname == "f32") {
Expand All @@ -449,30 +452,30 @@ void process_shaders() {
#if defined(GGML_VULKAN_COOPMAT2_GLSLC_SUPPORT)
if (tname == "f16") {
string_to_spv("flash_attn_f32_f16_" + tname, "flash_attn_cm2.comp",
merge_maps(base_dict, {{"Q_TYPE", "float"}, {"D_TYPE", "float"}, {"ACC_TYPE", acctype}}), true, false, true, f16acc);
merge_maps(fa_base_dict, {{"Q_TYPE", "float"}, {"D_TYPE", "float"}}), true, false, true, f16acc);
} else {
std::string data_a_key = "DATA_A_" + to_uppercase(tname);
string_to_spv("flash_attn_f32_f16_" + tname, "flash_attn_cm2.comp",
merge_maps(base_dict, {{data_a_key, "1"}, {"Q_TYPE", "float"}, {"D_TYPE", "float"}, {"ACC_TYPE", acctype}, {"DEQUANTFUNC", "dequantFunc"+to_uppercase(tname) }, {"BLOCK_SIZE", "QUANT_K_"+to_uppercase(tname) }}), true, false, true, f16acc);
merge_maps(fa_base_dict, {{data_a_key, "1"}, {"Q_TYPE", "float"}, {"D_TYPE", "float"}, {"DEQUANTFUNC", "dequantFunc"+to_uppercase(tname) }, {"BLOCK_SIZE", "QUANT_K_"+to_uppercase(tname) }}), true, false, true, f16acc);
}
#endif
#if defined(GGML_VULKAN_COOPMAT_GLSLC_SUPPORT)
if (tname == "f16") {
string_to_spv("flash_attn_f32_f16_" + tname, "flash_attn_cm1.comp",
merge_maps(base_dict, {{"Q_TYPE", "float"}, {"D_TYPE", "float"}, {"ACC_TYPE", acctype}, {"ACC_TYPEV4", acctypev4}, {"COOPMAT", "1"}}), true, true, false, f16acc);
merge_maps(fa_base_dict, {{"Q_TYPE", "float"}, {"D_TYPE", "float"}, {"COOPMAT", "1"}}), true, true, false, f16acc);
} else if (tname == "q4_0" || tname == "q8_0") {
std::string data_a_key = "DATA_A_" + to_uppercase(tname);
string_to_spv("flash_attn_f32_f16_" + tname, "flash_attn_cm1.comp",
merge_maps(base_dict, {{data_a_key, "1"}, {"Q_TYPE", "float"}, {"D_TYPE", "float"}, {"ACC_TYPE", acctype}, {"ACC_TYPEV4", acctypev4}, {"BLOCK_SIZE", "QUANT_K_"+to_uppercase(tname)}, {"COOPMAT", "1"}}), true, true, false, f16acc);
merge_maps(fa_base_dict, {{data_a_key, "1"}, {"Q_TYPE", "float"}, {"D_TYPE", "float"}, {"BLOCK_SIZE", "QUANT_K_"+to_uppercase(tname)}, {"COOPMAT", "1"}}), true, true, false, f16acc);
}
#endif
if (tname == "f16") {
string_to_spv("flash_attn_f32_f16_" + tname, "flash_attn.comp",
merge_maps(base_dict, {{"Q_TYPE", "float"}, {"D_TYPE", "float"}, {"ACC_TYPE", acctype}}), true, false, false, f16acc);
merge_maps(fa_base_dict, {{"Q_TYPE", "float"}, {"D_TYPE", "float"}}), true, false, false, f16acc);
} else if (tname == "q4_0" || tname == "q8_0") {
std::string data_a_key = "DATA_A_" + to_uppercase(tname);
string_to_spv("flash_attn_f32_f16_" + tname, "flash_attn.comp",
merge_maps(base_dict, {{data_a_key, "1"}, {"Q_TYPE", "float"}, {"D_TYPE", "float"}, {"ACC_TYPE", acctype}, {"BLOCK_SIZE", "QUANT_K_"+to_uppercase(tname) }}), true, false, false, f16acc);
merge_maps(fa_base_dict, {{data_a_key, "1"}, {"Q_TYPE", "float"}, {"D_TYPE", "float"}, {"BLOCK_SIZE", "QUANT_K_"+to_uppercase(tname) }}), true, false, false, f16acc);
}
}
}
Expand Down