Skip to content

GitLab

"references/vscode:/vscode.git/clone" did not exist on "4bf6c6e4b4cf2dea9c7f9952aa30d5820ab8ae33"
Unverified Commit 2aba569a authored by Thomas Stocker's avatar Thomas Stocker Committed by GitHub
Browse files

Vulkan based on #9650 (#11835) 

* implement the vulkan C backend

* add support in gpu.go

* add support in gen_linux.sh

* it builds

* fix segfault

* fix compilation

* fix free memory monitor

* fix total memory monitor

* update gpu.go

* fix build

* fix check_perfmon len

* remove cap_get_bound check

* fix vulkan handle releasing

* fix build on federa 40

* fix vulkan on windows

* making amdgpu work on arm achitecutre with vulkan

* add x86_64 lines in VulkanGlobs and capLinuxGlobs

* add aarch64 lines in vulkanGlobs and capLinuxGlobs

* Fix variable name

* Add vulkan build patch from @jmorganca

* Sync vendored ggml to add Vulkan support

* Updated dockerfile

https://github.com/whyvl/ollama-vulkan/issues/7#issuecomment-2660836871

Signed-off-by: default avatarVadim Grinco <vadim@grinco.eu>

* Installing rocm library
Signed-off-by: default avatarVadim Grinco <vadim@grinco.eu>

* This version works well

built based on this: https://github.com/whyvl/ollama-vulkan/issues/7#issuecomment-2660836871

Signed-off-by: default avatarVadim Grinco <vadim@grinco.eu>

* Applied 00-fix-vulkan-building.patch

Work done by McBane87 here: https://github.com/whyvl/ollama-vulkan/issues/7#issuecomment-2660836871

Signed-off-by: default avatarVadim Grinco <vadim@grinco.eu>

* Fixed the "detached head" issues
Signed-off-by: default avatarVadim Grinco <vadim@grinco.eu>

* Merged in the right direction
Signed-off-by: default avatarVadim Grinco <vadim@grinco.eu>

* Merging the latest stable (#2)

* Applied 00-fix-vulkan-building.patch

* Implemented vulkan backend based on the work done by whyvl, Dts0, McBane87 and others

Tested on AMD Ryzen 7 8845HS w/ Radeon 780M Graphics with ROCm disabled

```
[GIN-debug] POST   /v1/chat/completions      --> github.com/ollama/ollama/server.(*Server).ChatHandler-fm (6 handlers)
[GIN-debug] POST   /v1/completions           --> github.com/ollama/ollama/server.(*Server).GenerateHandler-fm (6 handlers)
[GIN-debug] POST   /v1/embeddings            --> github.com/ollama/ollama/server.(*Server).EmbedHandler-fm (6 handlers)
[GIN-debug] GET    /v1/models                --> github.com/ollama/ollama/server.(*Server).ListHandler-fm (6 handlers)
[GIN-debug] GET    /v1/models/:model         --> github.com/ollama/ollama/server.(*Server).ShowHandler-fm (6 handlers)
time=2025-03-11T13:00:40.793Z level=INFO source=gpu.go:199 msg="vulkan: load libvulkan and libcap ok"
time=2025-03-11T13:00:40.877Z level=INFO source=gpu.go:421 msg="error looking up vulkan GPU memory" error="device is a CPU"
time=2025-03-11T13:00:40.878Z level=WARN source=amd_linux.go:443 msg="amdgpu detected, but no compatible rocm library found.  Either install rocm v6, or follow manual install instructions at https://github.com/ollama/ollama/blob/main/docs/linux.md#manual-install"
time=2025-03-11T13:00:40.878Z level=WARN source=amd_linux.go:348 msg="unable to verify rocm library: no suitable rocm found, falling back to CPU"
time=2025-03-11T13:00:40.879Z level=INFO source=types.go:137 msg="inference compute" id=0 library=vulkan variant="" compute=1.3 driver=1.3 name="AMD Radeon Graphics (RADV GFX1103_R1)" total="15.6 GiB" available="15.6 GiB"
```

```
 # ollama run phi4:14b
>>> /set verbose
Set 'verbose' mode.
>>> how's it going?
Hello! I'm here to help you with any questions or tasks you have. How can I assist you today? 😊



total duration:       3.341959745s
load duration:        18.165612ms
prompt eval count:    15 token(s)
prompt eval duration: 475ms
prompt eval rate:     31.58 tokens/s
eval count:           26 token(s)
eval duration:        2.846s
eval rate:            9.14 tokens/s
>>>
```

* This is no longer needed
Signed-off-by: default avatarVadim Grinco <vadim@grinco.eu>

* Fixes SIGSEGV: segmentation violation running gemma3 models on ollama 0.6.0 #21

Patch provided by McBane87 on https://github.com/whyvl/ollama-vulkan/issues/21

Signed-off-by: default avatarVadim Grinco <vadim@grinco.eu>

* Applied 04-disable-mmap-vulkan.patch

From: https://github.com/whyvl/ollama-vulkan/issues/7#issuecomment-2660836871

Signed-off-by: default avatarVadim Grinco <vadim@grinco.eu>

* Pulled new upstream code for ggml-bulkan backend
Signed-off-by: default avatarVadim Grinco <vadim@grinco.eu>

* Merged latest ollama 0.6.2 and nasrally's Flash Attention patches (#5)

* readme: add Ellama to list of community integrations (#9800)

* readme: add screenpipe to community integrations (#9786)

* Add support for ROCm gfx1151 (#9773)

* conditionally enable parallel pipelines

* sample: make mutations in transforms explicit (#9743)

* updated minP to use early exit making use of sorted tokens

* ml/backend/ggml: allocate memory with malloc when loading model (#9822)

* runner: remove cache prompt flag from ollama runner (#9826)

We do not need to bypass the prompt caching in the ollama runner yet, as
only embedding models needed to bypass the prompt caching. When embedding
models are implemented they can skip initializing this cache completely.

* ollamarunner: Check for minBatch of context space when shifting

Models can specify that a group of inputs need to be handled a single
batch. However, context shifting didn't respect this and could trigger
a break anyways. In this case, we should instead trigger a context
shift earlier so that it occurs before the grouped batch.

Note that there still some corner cases:
 - A long prompt that exceeds the context window can get truncated
   in the middle of an image. With the current models, this will
   result in the model not recognizing the image at all, which is
   pretty much the expected result with truncation.
 - The context window is set less than the minimum batch size. The
   only solution to this is to refuse to load the model with these
   settings. However, this can never occur with current models and
   default settings.

Since users are unlikely to run into these scenarios, fixing them is
left as a follow up.

* Applied latest patches from McBane87

See this for details: https://github.com/whyvl/ollama-vulkan/issues/7#issuecomment-2708820861

Signed-off-by: default avatarVadim Grinco <vadim@grinco.eu>

* Add ability to enable flash attention on vulkan (#4

)

* discover: add flash attention handling for vulkan
* envconfig: fix typo in config.go

As part of the process some code was refactored and I added a new field
FlashAttention to GpuInfo since the previous solution didn't allow for a
granular check via vulkan extensions. As a side effect, this now allows
for granular per-device FA support checking in other places

---------
Signed-off-by: default avatarVadim Grinco <vadim@grinco.eu>
Co-authored-by: default avatarzeo <108888572+zeozeozeo@users.noreply.github.com>
Co-authored-by: default avatarLouis Beaumont <louis.beaumont@gmail.com>
Co-authored-by: default avatarDaniel Hiltgen <dhiltgen@users.noreply.github.com>
Co-authored-by: default avatarMichael Yang <mxyng@pm.me>
Co-authored-by: default avatarParth Sareen <parth.sareen@ollama.com>
Co-authored-by: default avatarJeffrey Morgan <jmorganca@gmail.com>
Co-authored-by: default avatarBruce MacDonald <brucewmacdonald@gmail.com>
Co-authored-by: default avatarJesse Gross <jesse@ollama.com>
Co-authored-by: default avatarNikita <50599445+nasrally@users.noreply.github.com>

* Revert Readme changes

* Revert

* Revert changes in amd_linux.go

* Revert changes in amd_linux.go

* Remove flashattention setting gpu.go

* Revert whitespace changes in gpu.go

* Revert changes in transforms_test.go

* Revert changes in runner.go

* Revert changes in Makefile.sync

* Revert some unintented changes in Dockerfile

* Revert vulkan copy changes in Dockerfile

* Update Vulkan Code to de4c07f93783a1a96456a44dc16b9db538ee1618

* Fixed duplicate sync in ggml.go

* Revert changes in ggml.go

* Revert chnages in ggml.go

* enable falsh attention on vulkan

* revert remove parenthesis

* fixed flash attention logic enabling

* vk_check_flash_attention 0 means supported

* Update gpu.go

* Add vulkan to Windows Build script

* Remove commented out code

* Enable Vulkan Flash attention in FlashAttentionSupported

* Fix logging

* Update Vulkan backend to e54d41befcc1575f4c898c5ff4ef43970cead75f

* Removed libcap related code

libcap is not directly related to Vulkan and should be added by its own PR. It adds additional library dependencies for building and also requires users to run setcap or run ollama as root, which is not ideal for easy use

* Fix Unit Test (Add Vulkan Library)

* Add vulkan to TestHomogeneousGPUs
Test

* vulkan: get GPU ID (ollama v0.11.5)
Signed-off-by: default avatarXiaodong Ye <xiaodong.ye@mthreads.com>

* disable mmap for vulkan

* Reduce Changes remove TestHomogeneousGPUs (doesn't exist on master)

* Update vulkan version to the version used in llama.cpp

* rename gpu patch to correct number

* added Vulkan API to get correct Device UUID

current UUID from pipelineCacheUUID does not match CUDA

* Fix GPU ID Patch

* Remove Code not in llama.cpp

* modified UUID code inside ggml

* Fix Patch

* Copied minimal definition from vulkan header

* Fix compile error in Mac

Metal is preferred so we're disabling Vulkan for now

* Removed unused code

Fix linter error in CI

* Fix patches apply

* fixing lint error

* Removed unneeded function call

Somehow removing this call fixed the crashing when Vulkan header was removed

* added missing NL

* Fixed missing members in Vulkan header

also added zero clear for some structs

* Fixed wrong structure ID

* Fixed Vulkan header

More aligned with official header definition now

* buildvulkanAsSeperateFunction

* Vulkan on Windows Test

* temporarly comment out gate to run windows task

* use temporarly windows-latest for build

* Commenting out other presets to build vulkan

* reenable cpu

* commenting out error action stop

* temporarly commenting out rocm

* set vulkan path

* comment out cude for faster turnaround

* correct vulkan install

* correct vulkan silent install

* fixed install command

* revert debugging changes (vulkan builds on windows)

* revert windows-latest

* trying to build vulkan for linux

* temporarly disable cuda and rocm

* try again linux build

* fix version

* trying to fix

* trying again

* trying again

* fix version

* fixed vulkan-sdk name

* try again

* trying again

* try without version number

* try again

* add some more extra

* trying to use version 1.4.313

* revert debugging changes

* Filter out already supported gpus

* revert debug code

* Use runners for GPU discovery

This revamps how we discover GPUs in the system by leveraging the Ollama
runner.  This should eliminate inconsistency between our GPU discovery and the
runners capabilities at runtime, particularly for cases where we try to filter
out unsupported GPUs.  Now the runner does that implicitly based on the actual
device list.  In some cases free VRAM reporting can be unreliable which can
leaad to scheduling mistakes, so this also includes a patch to leverage more
reliable VRAM reporting libraries if available.

Automatic workarounds have been removed as only one GPU leveraged this, which
is now documented. This GPU will soon fall off the support matrix with the next
ROCm bump.

Additional cleanup of the scheduler and discovery packages can be done in the
future once we have switched on the new memory management code, and removed
support for the llama runner.

* timing info for runner

* WIP - wire up Vulkan with the new engine based discovery

Not a complete implementation - free VRAM is better, but not accurate on
windows

* fix - trust the library paths from discovery when starting runner

* fix index bug

* fix vulkan ids to be underlying

* fix - give bootstrapping more time on slow systems

* Test if Vulkan device is supported

* vk_check_flash_attention is not needed (coompat2 coopmapt and scalar implementation exist)

* Handle GGML_VK_VISIBLE_DEVICES

* ask for supported first

* win: fix CPU query buffer handling

Try in a short loop until we get the size right.

* test: harden integration tests for slow start

If the server takes a while to start up, block
tests from starting until it's online to avoid
setting large timeouts in individual test cases.

* gofumpt fix

* fix build

* merge fixes

* merge fixes

* fixed build

* merge fixes

* fixing build

* fixed build

* fixed formatting

* fixed build

* fix vulkan gpu id patch

* sync llama.cpp vulkan code

* update build windows script

* merge fixes

* fix format

* fixed vulkan casing

* handle igpu as gpu

* improve case

* print out unknown library

* rturn Vulkan for vulkan library

* Revert "rturn Vulkan for vulkan library"

This reverts commit 690461a12fd5e93295d174c97edefb2bc33285b1.

* fixed patch number

* return Library Name

* remvoe debug code

* return integrated in vulkan backend

* Return pci Properties

* update patch

* directly get pci proeprties without parsing

* workaround for filtering devices. Correct way is to have a LibraryPosition Parameter in the deviceInfo

* Revert "directly get pci proeprties without parsing"

This reverts commit 8e0624851f5ed7d9f74518f574dfb422e4dd4dc2.

* Set FilteredID for Environment Filtering

* ROCm Library is named ROCm

* revert changes in patch

* Create 0028-vulkan-pci-and-memory.patch

* vulkan memory patch

* casing fix

* Add more pci properties

* Added better memory management

* Added better memory managament

* fixed patch

* Fixed patch

* FilterID creation group by library

* filter out vulkan supported by other gpu

* fixing deviceid compare

* Vulkan Fix FA coopmat1 invalid array indexing

* Use everywhere the same Vulkan Version 1.4.321.1

* Remove unneeded patch

* vulkan update

* sync vulkan glsl files

* only use for vulkan the filteredid (numeric device number)

* simplify code

---------
Signed-off-by: default avatarVadim Grinco <vadim@grinco.eu>
Signed-off-by: default avatarXiaodong Ye <xiaodong.ye@mthreads.com>
Co-authored-by: default avatarpufferffish <github@bandersnatch.anonaddy.com>
Co-authored-by: KOISHI KOMEIJI FROM TOUHOU 11 <fuck>
Co-authored-by: default avatarDSLstandard <qgeneral35@gmail.com>
Co-authored-by: default avatarpufferffish <me@windtfw.com>
Co-authored-by: default avataryeongbba <yeongmo.lee@logpresso.com>
Co-authored-by: default avatartomaThomas <tomathomas@mailbox.org>
Co-authored-by: default avatarAntoine Viallon <antoine@lesviallon.fr>
Co-authored-by: default avatarVadim Grinco <vadim@grinco.eu>
Co-authored-by: default avatarzeo <108888572+zeozeozeo@users.noreply.github.com>
Co-authored-by: default avatarLouis Beaumont <louis.beaumont@gmail.com>
Co-authored-by: default avatarDaniel Hiltgen <dhiltgen@users.noreply.github.com>
Co-authored-by: default avatarMichael Yang <mxyng@pm.me>
Co-authored-by: default avatarParth Sareen <parth.sareen@ollama.com>
Co-authored-by: default avatarJeffrey Morgan <jmorganca@gmail.com>
Co-authored-by: default avatarBruce MacDonald <brucewmacdonald@gmail.com>
Co-authored-by: default avatarJesse Gross <jesse@ollama.com>
Co-authored-by: default avatarNikita <50599445+nasrally@users.noreply.github.com>
Co-authored-by: default avatarMasato Nakasaka <masato.nakasaka@intel.com>
Co-authored-by: default avatarXiaodong Ye <xiaodong.ye@mthreads.com>
Co-authored-by: default avatarDaniel Hiltgen <daniel@ollama.com>
parent fd8aa947
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 1946 additions and 0 deletions
ml/backend/ggml/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_base.glsl 0 → 100644
View file @ 2aba569a
layout(local_size_x_id = 0, local_size_y = 1, local_size_z = 1) in;
layout (constant_id = 0) const uint32_t WorkGroupSize = 128;
layout (constant_id = 1) const uint32_t Br = 1;
layout (constant_id = 2) const uint32_t Bc = 32;
layout (constant_id = 3) const uint32_t HSK = 32;
layout (constant_id = 4) const uint32_t HSV = 32;
layout (constant_id = 5) const uint32_t Clamp = 0;
layout (constant_id = 6) const uint32_t D_split = 16;
// Round up head sizes to a multiple of 16, for coopmat1/coopmat2 paths
const uint32_t HSK_pad = (HSK + 15) & ~15;
const uint32_t HSV_pad = (HSV + 15) & ~15;
const bool KV_bounds_check = Clamp != 0;
layout (push_constant) uniform parameter {
uint32_t N;
uint32_t KV;
uint32_t ne1;
uint32_t ne2;
uint32_t ne3;
uint32_t neq2;
uint32_t neq3;
uint32_t nek2;
uint32_t nek3;
uint32_t nev2;
uint32_t nev3;
uint32_t nem1;
uint32_t nem2;
uint32_t nem3;
uint32_t nb01;
uint32_t nb02;
uint32_t nb03;
uint32_t nb11;
uint32_t nb12;
uint32_t nb13;
uint32_t nb21;
uint32_t nb22;
uint32_t nb23;
float scale;
float max_bias;
float logit_softcap;
uint32_t mask_n_head_log2;
float m0;
float m1;
uint32_t gqa_ratio;
uint32_t split_kv;
uint32_t k_num;
} p;
#define SINK_ENABLE_BIT (1<<24)
#define MASK_ENABLE_BIT (1<<16)
#define N_LOG2_MASK 0xFFFF
layout (binding = 4) readonly buffer S {float data_s[];};
layout (binding = 5) writeonly buffer O {D_TYPE data_o[];};
#if defined(A_TYPE_PACKED16)
#define BINDING_IDX_K 0
#define BINDING_IDX_V 1
layout (binding = 1) readonly buffer K_PACKED16 {A_TYPE_PACKED16 k_data_packed16[];} k_packed;
layout (binding = 2) readonly buffer V_PACKED16 {A_TYPE_PACKED16 v_data_packed16[];} v_packed;
#endif
#if defined(DATA_A_Q4_0)
#define BLOCK_BYTE_SIZE 18
vec4 dequantize4(uint ib, uint iqs, uint a_offset, uint binding_idx) {
if (binding_idx == BINDING_IDX_K) {
uint vui_lo = uint(k_packed.k_data_packed16[a_offset + ib].qs[(iqs & 0xF) / 2 + 0]);
uint vui_hi = uint(k_packed.k_data_packed16[a_offset + ib].qs[(iqs & 0xF) / 2 + 1]);
uint shift = (iqs & 0x10) >> 2;
vui_lo >>= shift;
vui_hi >>= shift;
return float(k_packed.k_data_packed16[a_offset + ib].d) * (vec4(vui_lo & 0xF, (vui_lo >> 8) & 0xF, vui_hi & 0xF, (vui_hi >> 8) & 0xF) - 8.0f);
} else {
uint vui_lo = uint(v_packed.v_data_packed16[a_offset + ib].qs[(iqs & 0xF) / 2 + 0]);
uint vui_hi = uint(v_packed.v_data_packed16[a_offset + ib].qs[(iqs & 0xF) / 2 + 1]);
uint shift = (iqs & 0x10) >> 2;
vui_lo >>= shift;
vui_hi >>= shift;
return float(v_packed.v_data_packed16[a_offset + ib].d) * (vec4(vui_lo & 0xF, (vui_lo >> 8) & 0xF, vui_hi & 0xF, (vui_hi >> 8) & 0xF) - 8.0f);
}
}
#endif
#if defined(DATA_A_Q8_0)
#define BLOCK_BYTE_SIZE 34
vec4 dequantize4(uint ib, uint iqs, uint a_offset, uint binding_idx) {
if (binding_idx == BINDING_IDX_K) {
const i8vec2 v0 = unpack8(int32_t(k_packed.k_data_packed16[a_offset + ib].qs[iqs / 2])).xy; // vec4 used due to #12147
const i8vec2 v1 = unpack8(int32_t(k_packed.k_data_packed16[a_offset + ib].qs[iqs / 2 + 1])).xy;
return float(k_packed.k_data_packed16[a_offset + ib].d) * vec4(v0.x, v0.y, v1.x, v1.y);
} else {
const i8vec2 v0 = unpack8(int32_t(v_packed.v_data_packed16[a_offset + ib].qs[iqs / 2])).xy; // vec4 used due to #12147
const i8vec2 v1 = unpack8(int32_t(v_packed.v_data_packed16[a_offset + ib].qs[iqs / 2 + 1])).xy;
return float(v_packed.v_data_packed16[a_offset + ib].d) * vec4(v0.x, v0.y, v1.x, v1.y);
}
}
#endif
#define CEIL_DIV(a, b) (((a) + (b) - 1) / (b))
// Store column zero. This is used to save per-row m and L values for split_k.
ACC_TYPE perElemOpStoreCol0(const in uint32_t r, const in uint32_t c, const in ACC_TYPE elem, const in uint32_t o_offset, const in uint32_t iq2, const in uint32_t N)
{
if (r < N && c == 0) {
uint32_t offset = iq2 + r;
data_o[o_offset + offset] = D_TYPE(elem);
}
return elem;
}
// Load the slope matrix, indexed by Q's dimension 2.
ACC_TYPE perElemOpComputeSlope(const in uint32_t r, const in uint32_t c, const in ACC_TYPE elem, const in uint32_t iq2)
{
const uint32_t h = iq2 + (r % p.gqa_ratio);
uint32_t n_head_log2 = p.mask_n_head_log2 & N_LOG2_MASK;
const ACC_TYPE base = ACC_TYPE(h < n_head_log2 ? p.m0 : p.m1);
const int exph = int(h < n_head_log2 ? h + 1 : 2*(h - n_head_log2) + 1);
return ACC_TYPE(pow(base, ACC_TYPE(exph)));
}
// Load the sink value, indexed by Q's dimension 2.
ACC_TYPE perElemOpGetSink(const in uint32_t r, const in uint32_t c, const in ACC_TYPE elem, const in uint32_t iq2)
{
const uint32_t h = iq2 + (r % p.gqa_ratio);
return ACC_TYPE(data_s[h]);
}
uint32_t i, N, KV, split_k_index, Tr, start_j, end_j,
iq2, iq3, rk2, rk3, rv2, rv3, ik2, ik3, iv2, iv3,
q_stride, k_stride, v_stride, m_stride;
void init_indices()
{
N = p.N;
KV = p.KV;
i = gl_WorkGroupID.x;
split_k_index = 0;
if (p.k_num > 1) {
i = 0;
split_k_index = gl_WorkGroupID.x;
}
Tr = CEIL_DIV(N, Br);
start_j = split_k_index * p.split_kv / Bc;
end_j = CEIL_DIV(min(KV, (split_k_index + 1) * p.split_kv), Bc);
// When not using grouped query attention, all rows share the same iq2, equal to gl_WorkGroupID.y.
// When using grouped query attention, each workgroup does gqa_ratio consecutive values of iq2.
iq2 = gl_WorkGroupID.y * p.gqa_ratio;
iq3 = gl_WorkGroupID.z;
// broadcast factors
rk2 = p.neq2/p.nek2;
rk3 = p.neq3/p.nek3;
rv2 = p.neq2/p.nev2;
rv3 = p.neq3/p.nev3;
// k indices
ik3 = iq3 / rk3;
ik2 = iq2 / rk2;
// v indices
iv3 = iq3 / rv3;
iv2 = iq2 / rv2;
// nb?1 are already divided by the type size and are in units of elements.
// When using grouped query attention, Q is indexed by iq2, so the stride
// should be nb02 (which is in bytes).
q_stride = p.gqa_ratio > 1 ? (p.nb02 / 4) : p.nb01;
k_stride = p.nb11;
v_stride = p.nb21;
// When using grouped query attention, all rows use the same mask (stride 0).
// "p.gqa_ratio >> 16" is just a roundabout way of writing zero
// that prevents the compiler from folding the "&" through the select
// and breaking the alignment detection.
m_stride = (p.gqa_ratio > 1) ? (p.gqa_ratio >> 16) : KV;
}
ml/backend/ggml/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_split_k_reduce.comp 0 → 100644
View file @ 2aba569a
#version 450
#extension GL_EXT_control_flow_attributes : enable
layout(constant_id = 0) const uint BLOCK_SIZE = 32;
layout(local_size_x_id = 0, local_size_y = 1, local_size_z = 1) in;
layout (binding = 0) readonly buffer A {float data_a[];};
layout (binding = 1) readonly buffer B {float data_s[];};
layout (binding = 2) writeonly buffer D {float data_d[];};
layout (push_constant) uniform parameter {
uint D;
uint N;
uint ne3;
uint k_num;
uint sinks;
} p;
shared float tmpsh[BLOCK_SIZE];
void main() {
// Each workgroup handles a row
const uint n = gl_WorkGroupID.x;
const uint tid = gl_LocalInvocationID.x;
const uint iq3 = gl_WorkGroupID.z;
uint D = p.D;
uint N = p.N;
uint k_num = p.k_num;
uint l_offset = D * N * p.ne3 * k_num + N * iq3 * k_num * 2 + n;
uint m_offset = D * N * p.ne3 * k_num + N * iq3 * k_num * 2 + N + n;
uint lm_stride = N * 2;
// Compute the max m value for the row
float m_max = -1.0/0.0;
for (uint k = 0; k + tid < k_num; k += BLOCK_SIZE) {
float m = data_a[m_offset + (k + tid) * lm_stride];
m_max = max(m_max, m);
}
// reduce across the workgroup
tmpsh[tid] = m_max;
barrier();
[[unroll]] for (uint s = BLOCK_SIZE/2; s > 0; s >>= 1) {
if (tid < s) {
m_max = max(m_max, tmpsh[tid + s]);
tmpsh[tid] = m_max;
}
barrier();
}
m_max = tmpsh[0];
barrier();
// Compute L based on m_max
float L = 0;
for (uint k = 0; k + tid < k_num; k += BLOCK_SIZE) {
float l = data_a[l_offset + (k + tid) * lm_stride];
float m = data_a[m_offset + (k + tid) * lm_stride];
L += exp(m - m_max) * l;
}
// reduce across the workgroup
tmpsh[tid] = L;
barrier();
[[unroll]] for (uint s = BLOCK_SIZE/2; s > 0; s >>= 1) {
if (tid < s) {
L += tmpsh[tid + s];
tmpsh[tid] = L;
}
barrier();
}
L = tmpsh[0];
float sink;
if (p.sinks != 0) {
sink = data_s[n];
float ms = 1.0f;
float vs = 1.0f;
if (sink > m_max) {
ms = exp(m_max - sink);
} else {
vs = exp(sink - m_max);
}
L = L*ms + vs;
}
L = 1.0 / L;
// D dimension is split across workgroups in the y dimension
uint d = tid + gl_WorkGroupID.y * BLOCK_SIZE;
// Scale and sum the O contributions based on m_max and store the result to memory
if (d < D) {
float O = 0.0;
[[unroll]] for (uint k = 0; k < k_num; ++k) {
uint o_offset = D * N * (k + iq3 * k_num) + D * n + d;
float m = data_a[m_offset + k * lm_stride];
O += exp(m - m_max) * data_a[o_offset];
}
if (p.sinks != 0) {
if (sink > m_max) {
float ms = 1.0f;
ms = exp(m_max - sink);
O *= ms;
}
}
O *= L;
const float FLT_MAX = uintBitsToFloat(0x7F7FFFFF);
O = clamp(O, -FLT_MAX, FLT_MAX);
data_d[iq3 * D * N + D * n + d] = O;
}
}
ml/backend/ggml/ggml/src/ggml-vulkan/vulkan-shaders/geglu.comp 0 → 100644
View file @ 2aba569a
#version 450
#include "glu_head.glsl"
const float GELU_COEF_A = 0.044715f;
const float SQRT_2_OVER_PI = 0.79788456080286535587989211986876f;
float op(float a, float b) {
const float val = SQRT_2_OVER_PI*a*(1.0f + GELU_COEF_A*a*a);
return 0.5f*a*(2.0f - 2.0f / (exp(2 * val) + 1)) * b;
}
#include "glu_main.glsl"
ml/backend/ggml/ggml/src/ggml-vulkan/vulkan-shaders/geglu_erf.comp 0 → 100644
View file @ 2aba569a
#version 450
#include "glu_head.glsl"
// based on Abramowitz and Stegun formula 7.1.26 or similar Hastings' approximation
// ref: https://www.johndcook.com/blog/python_erf/
const float p_erf = 0.3275911f;
const float a1_erf = 0.254829592f;
const float a2_erf = -0.284496736f;
const float a3_erf = 1.421413741f;
const float a4_erf = -1.453152027f;
const float a5_erf = 1.061405429f;
const float SQRT_2_INV = 0.70710678118654752440084436210484f;
float op(float a, float b) {
const float a_div_sqr2 = a * SQRT_2_INV;
const float sign_x = sign(a_div_sqr2);
const float x = abs(a_div_sqr2);
const float t = 1.0f / (1.0f + p_erf * x);
const float y = 1.0f - (((((a5_erf * t + a4_erf) * t) + a3_erf) * t + a2_erf) * t + a1_erf) * t * exp(-x * x);
const float erf_approx = sign_x * y;
return 0.5f * a * (1.0f + erf_approx) * b;
}
#include "glu_main.glsl"
ml/backend/ggml/ggml/src/ggml-vulkan/vulkan-shaders/geglu_quick.comp 0 → 100644
View file @ 2aba569a
#version 450
#include "glu_head.glsl"
const float GELU_QUICK_COEF = -1.702f;
float op(float a, float b) {
return a * (1.0f / (1.0f + exp(GELU_QUICK_COEF * a))) * b;
}
#include "glu_main.glsl"
ml/backend/ggml/ggml/src/ggml-vulkan/vulkan-shaders/gelu.comp 0 → 100644
View file @ 2aba569a
This diff is collapsed.
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment