Skip to content

vulkan: Warptile tuning for Intel Xe2/Xe3 #18178

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

Open
virajwad wants to merge 8 commits into
base: master
Choose a base branch
from
+28 −9
Open

vulkan: Warptile tuning for Intel Xe2/Xe3 #18178

Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
0f82d0a
modify warptile tuning for xe3
virajwad Dec 15, 2025
c908711
intel vendor check w/ coopmat support
virajwad Dec 17, 2025
3441283
fix back formatting
virajwad Dec 17, 2025
6b7f1e8
fix formatting change 2
virajwad Dec 17, 2025
0cfa616
move intel check to chip specific tuning part
virajwad Dec 18, 2025
c40396a
Change to support both windows and linux
virajwad Dec 20, 2025
dc14792
modify m_warptile to l_warptile for intel
virajwad Dec 29, 2025
328116e
modify warptile tuning for bf16 matmuls to fix regression (m_warptile…
virajwad Dec 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
37 changes: 28 additions & 9 deletions ggml/src/ggml-vulkan/ggml-vulkan.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2838,18 +2838,25 @@ static void ggml_vk_load_shaders(vk_device& device) {
m_warptile_mmqid_int_k = { 128, 64, 64, 32, mul_mat_subgroup_size_16, 32, 1, 2, 2, 1, mul_mat_subgroup_size_16 };
s_warptile_mmqid_int_k = { mul_mat_subgroup_size_32, 32, 32, 32, 32, 32, 1, 2, 1, 1, mul_mat_subgroup_size_16 };

l_mmq_wg_denoms = l_wg_denoms = { 128, 128, 1 };
m_mmq_wg_denoms = m_wg_denoms = { 64, 64, 1 };
s_mmq_wg_denoms = s_wg_denoms = { 32, 32, 1 };
l_align = 128;
m_align = 64;
s_align = 32;

// chip specific tuning
if ((device->architecture == AMD_GCN) && (device->driver_id != vk::DriverId::eAmdProprietary)) {
m_warptile_mmq = m_warptile_mmq_int = { 256, 64, 64, 32, 16, 16, 2, 2, 2, 1, 16 };
m_warptile_mmqid = m_warptile_mmqid_int = { 256, 64, 64, 32, 16, 16, 2, 2, 2, 1, 16 };
}

l_mmq_wg_denoms = l_wg_denoms = {128, 128, 1 };
m_mmq_wg_denoms = m_wg_denoms = { 64, 64, 1 };
s_mmq_wg_denoms = s_wg_denoms = { 32, 32, 1 };
l_align = 128;
m_align = 64;
s_align = 32;
Copy link
Collaborator

@0cc4m 0cc4m Jan 5, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't see a reason this needs to be set earlier, so please leave it as is.

else if (device->vendor_id == VK_VENDOR_ID_INTEL) {
if (device->coopmat_support && device->architecture == INTEL_XE2) {
Copy link
Collaborator

@0cc4m 0cc4m Jan 5, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Let's stick to one if until we actually have more Intel cases that would justify a secondary branch.

// Xe2/Xe3 with coopmat enabled - warptile performance tuning
l_warptile = { 512, 128, 128, 16, subgroup_size_8, 32, 2, tm_m, tn_m, tk_m, subgroup_size_8 };
l_warptile_mmq = { 512, 128, 128, 32, subgroup_size_8, 32, 2, tm_m, tn_m, tk_m, subgroup_size_8 };
}
}

for (uint32_t i = 0; i < GGML_TYPE_COUNT; ++i) {
ggml_type t = (ggml_type)i;
Expand Down Expand Up @@ -3521,6 +3528,13 @@ static void ggml_vk_load_shaders(vk_device& device) {
m_wg_denoms = { 64, 64, 1 };
s_wg_denoms = { 32, 32, 1 };

if (device->vendor_id == VK_VENDOR_ID_INTEL) {
if (device->architecture == INTEL_XE2) {
Copy link
Collaborator

@0cc4m 0cc4m Jan 5, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

See above.

// Xe2/Xe3 - bf16 warptile performance tuning
l_warptile = { 512, 128, 128, 16, subgroup_size_8, 32, 2, 4, 4, 1, subgroup_size_8 };
}
}

CREATE_MM(GGML_TYPE_BF16, pipeline_matmul_bf16, matmul_bf16, , wg_denoms, warptile, vk_mat_mat_push_constants, 3, , 0);
CREATE_MM(GGML_TYPE_BF16, pipeline_matmul_id_bf16, matmul_id_bf16, , wg_denoms, warptile, vk_mat_mat_id_push_constants, 4, _id, 0);
}
Expand Down Expand Up @@ -4868,10 +4882,15 @@ static vk_device ggml_vk_get_device(size_t idx) {
#ifndef GGML_VULKAN_RUN_TESTS
case VK_VENDOR_ID_AMD:
case VK_VENDOR_ID_INTEL:
device->mul_mat_l[i] = false;
if (!device->coopmat_support || device->architecture != INTEL_XE2) {
device->mul_mat_l[i] = false;
device->mul_mat_id_l[i] = false;
} else {
device->mul_mat_l[i] = true; // if coopmat & XE2+, allow large matmul warptile config for Intel
device->mul_mat_id_l[i] = true;
}
Copy link
Collaborator

@0cc4m 0cc4m Jan 5, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Since we're now diverging between Intel and AMD, please copy the original AMD code to the top and separate them from Intel.

device->mul_mat_m[i] = true;
device->mul_mat_s[i] = true;
device->mul_mat_id_l[i] = false;
device->mul_mat_id_m[i] = true;
device->mul_mat_id_s[i] = true;
break;
Expand Down
Loading