armsx/ARMSX3
0
0
Fork 0
mirror of https://github.com/ARMSX2/ARMSX3.git synced 2026-08-24 16:58:52 -07:00
Code Issues Packages Projects Releases Wiki Activity

VK: allow the compute work group size to be overridden for benchmarking

Browse Source 
The mobile branch of the per-vendor group size table picks 32 by falling
through to the NVIDIA case. Adreno and Mali both run 64-wide waves, so 32
plausibly leaves half of each wave idle -- but that is reasoning, not a
measurement, and guessing wrong here costs performance silently.

Read ARMSX3_CS_GROUP_SIZE, which driver_env.txt already plumbs through
setenv(), so both candidates can be compared on one build without a settings
field or a second APK. Powers of two only, clamped to maxComputeWorkGroupSize
and maxComputeWorkGroupInvocations, because an over-large local_size_x fails
shader compilation rather than validation. The default is unchanged; this only
makes the question answerable.
This commit is contained in:
jpolo1224
2026-08-20 16:08:20 -04:00
parent 5d71742da9
commit cbee3cd44b
1 changed files with 23 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
rpcs3/Emu/RSX/VK/VKCompute.cpp
+23
View File
@@ -1,4 +1,5 @@
#include "VKCompute.h"
#include <cstdlib>
#include "Emu/RSX/rsx_profiler.h"
#include "VKHelpers.h"
#include "VKRenderPass.h"
@@ -97,6 +98,28 @@ namespace vk
const auto& gpu = vk::g_render_device->gpu();
max_invocations_x = gpu.get_limits().maxComputeWorkGroupCount[0];
// Every group size above is a per-vendor guess, and the mobile one has never been
// measured on a phone. This override exists so the candidates can be compared on a
// single build -- set ARMSX3_CS_GROUP_SIZE in driver_env.txt -- rather than
// shipping another guess. Powers of two only, and never past what the device
// accepts: an over-large local_size_x fails shader compilation, not validation.
if (const char* const env = std::getenv("ARMSX3_CS_GROUP_SIZE"))
{
const auto& limits = gpu.get_limits();
const u32 ceiling = std::min(limits.maxComputeWorkGroupSize[0], limits.maxComputeWorkGroupInvocations);
const u64 requested = std::strtoull(env, nullptr, 10);
if (requested >= 1 && requested <= ceiling && (requested & (requested - 1)) == 0)
{
rsx_log.warning("cs: work group size overridden %u -> %u by ARMSX3_CS_GROUP_SIZE.", optimal_group_size, static_cast<u32>(requested));
optimal_group_size = static_cast<u32>(requested);
}
else
{
rsx_log.error("cs: ARMSX3_CS_GROUP_SIZE='%s' ignored; expected a power of two in [1, %u].", env, ceiling);
}
}
initialized = true;
}
}