Skip to content

[ET-VK] Update vulkan gpuinfo code for staging buffer changes #16664

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
GregoryComer wants to merge 2 commits into
base: main
Choose a base branch
from
+37 −9
Open

[ET-VK] Update vulkan gpuinfo code for staging buffer changes #16664

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
29cb4f6
Update vulkan gpuinfo code for staging buffer changes
GregoryComer Jan 16, 2026
ca9705a
Lint
GregoryComer Jan 16, 2026
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
15 changes: 12 additions & 3 deletions backends/vulkan/tools/gpuinfo/include/architecture.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,7 +40,8 @@ void reg_count(const App& app) {
uint32_t NITER;

auto bench = [&](uint32_t ngrp, uint32_t nreg) {
StagingBuffer buffer(context(), vkapi::kFloat, 1);
StagingBuffer buffer(
context(), vkapi::kFloat, 1, vkapi::CopyDirection::DEVICE_TO_HOST);
vkapi::PipelineBarrier pipeline_barrier{};

auto shader_name = "reg_count_" + std::to_string(nreg);
Expand Down Expand Up @@ -164,7 +165,11 @@ void warp_size(const App& app, const bool verbose = false) {
uint32_t NITER;

auto bench = [&](uint32_t nthread) {
StagingBuffer out_buf(context(), vkapi::kInt, app.nthread_logic);
StagingBuffer out_buf(
context(),
vkapi::kInt,
app.nthread_logic,
vkapi::CopyDirection::DEVICE_TO_HOST);
vkapi::PipelineBarrier pipeline_barrier{};

auto shader_name = "warp_size_physical";
Expand Down Expand Up @@ -224,7 +229,11 @@ void warp_size(const App& app, const bool verbose = false) {
// doesn't depend on kernel timing, so the extra wait time doesn't lead to
// inaccuracy.
auto bench_sm = [&](uint32_t nthread) {
StagingBuffer out_buf(context(), vkapi::kInt, app.nthread_logic);
StagingBuffer out_buf(
context(),
vkapi::kInt,
app.nthread_logic,
vkapi::CopyDirection::DEVICE_TO_HOST);
vkapi::PipelineBarrier pipeline_barrier{};

auto shader_name = "warp_size_scheduler";
Expand Down
20 changes: 16 additions & 4 deletions backends/vulkan/tools/gpuinfo/include/buffers.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,8 +35,13 @@ void buf_cacheline_size(const App& app) {
uint32_t NITER;

auto bench = [&](int stride) {
StagingBuffer in_buf(context(), vkapi::kFloat, BUF_SIZE);
StagingBuffer out_buf(context(), vkapi::kFloat, 1);
StagingBuffer in_buf(
context(),
vkapi::kFloat,
BUF_SIZE,
vkapi::CopyDirection::HOST_TO_DEVICE);
StagingBuffer out_buf(
context(), vkapi::kFloat, 1, vkapi::CopyDirection::DEVICE_TO_HOST);
vkapi::PipelineBarrier pipeline_barrier{};

auto shader_name = "buf_cacheline_size";
Expand Down Expand Up @@ -132,9 +137,16 @@ void _bandwidth(
// workgroups, once the size of the access excedes the workgroup width.
const uint32_t workgroup_width = local_x * NITER * NUNROLL;

StagingBuffer in_buf(context(), vkapi::kFloat, range / sizeof(float));
StagingBuffer in_buf(
context(),
vkapi::kFloat,
range / sizeof(float),
vkapi::CopyDirection::HOST_TO_DEVICE);
StagingBuffer out_buf(
context(), vkapi::kFloat, VEC_WIDTH * app.nthread_logic);
context(),
vkapi::kFloat,
VEC_WIDTH * app.nthread_logic,
vkapi::CopyDirection::DEVICE_TO_HOST);
vkapi::PipelineBarrier pipeline_barrier{};

auto shader_name = "buf_bandwidth_" + memtype_lower;
Expand Down
11 changes: 9 additions & 2 deletions backends/vulkan/tools/gpuinfo/include/textures.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -61,7 +61,11 @@ void tex_cacheline_concurr(const App& app) {
vTensor in_tensor =
api::vTensor(api::context(), sizes_nchw, vkapi::kFloat);

StagingBuffer out_buf(context(), vkapi::kFloat, TEXEL_WIDTH);
StagingBuffer out_buf(
context(),
vkapi::kFloat,
TEXEL_WIDTH,
vkapi::CopyDirection::DEVICE_TO_HOST);

vkapi::PipelineBarrier pipeline_barrier{};

Expand Down Expand Up @@ -174,7 +178,10 @@ void tex_bandwidth(const App& app) {
const uint32_t workgroup_width = local_x * NITER * NUNROLL;

StagingBuffer out_buf(
context(), vkapi::kFloat, VEC_WIDTH * app.nthread_logic);
context(),
vkapi::kFloat,
VEC_WIDTH * app.nthread_logic,
vkapi::CopyDirection::DEVICE_TO_HOST);
vkapi::PipelineBarrier pipeline_barrier{};

auto time = benchmark_on_gpu(shader_name, 10, [&]() {
Expand Down
Loading