fabi/Xenia-Canary
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[GPU] EVENT_WRITE_ZPD batched sample accumulation cvar

Browse Source
This commit is contained in:
goldislead
2026-04-03 22:17:43 -07:00
committed by Radosław Gliński
parent 7e39a7018f
commit 9c00ce9366
3 changed files with 21 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
src/xenia/gpu/gpu_flags.cc
View File

@@ -59,6 +59,11 @@ DEFINE_bool(
"when MSAA is used with fullscreen passes.",
"GPU");
DEFINE_bool(query_occlusion_batched, false,
"Increment the sample count total by a fixed amount on every "
"EVENT_WRITE_ZPD. This provides an approximate batched occlusion "
"query implementation for many titles.",
"GPU");
DEFINE_int32(query_occlusion_sample_lower_threshold, 80,
"If set to -1 no sample counts are written, games may hang. Else, "
"the sample count of every tile will be incremented on every "

2
src/xenia/gpu/gpu_flags.h
View File

@@ -26,6 +26,8 @@ DECLARE_bool(non_seamless_cube_map);
DECLARE_bool(half_pixel_offset);
DECLARE_bool(query_occlusion_batched);
DECLARE_int32(query_occlusion_sample_lower_threshold);
DECLARE_int32(query_occlusion_sample_upper_threshold);

15
src/xenia/gpu/pm4_command_processor_implement.h
View File

@@ -955,6 +955,7 @@ bool COMMAND_PROCESSOR::ExecutePacketType3_EVENT_WRITE_EXT(
}
static uint32_t samples = cvars::query_occlusion_sample_upper_threshold;
static uint32_t batched_samples = 0;
XE_NOINLINE
bool COMMAND_PROCESSOR::ExecutePacketType3_EVENT_WRITE_ZPD(
@@ -982,7 +983,19 @@ bool COMMAND_PROCESSOR::ExecutePacketType3_EVENT_WRITE_ZPD(
bool is_end_via_z_fail = pSampleCounts->ZFail_A == kQueryFinished &&
pSampleCounts->ZFail_B == kQueryFinished;
std::memset(pSampleCounts, 0, sizeof(xe_gpu_depth_sample_counts));
if (is_end_via_z_pass || is_end_via_z_fail) {
// Titles that use QueryBatch (4D5309B1, 4E4D0801) don't use an END result for
// every query. Each ISSUE snapshots the sample count into a new slot and
// recovers the total by looking at differences between slots.
// END detection isn't sufficient here.
if (cvars::query_occlusion_batched) {
// Mimic batched behavior by reporting a running count and writing it on
// every event.
const uint32_t step = std::max(uint32_t(1), samples);
batched_samples = std::min(batched_samples, UINT32_MAX - step) + step;
pSampleCounts->ZPass_A = batched_samples;
pSampleCounts->Total_A = batched_samples;
} else if (is_end_via_z_pass || is_end_via_z_fail) {
pSampleCounts->ZPass_A = samples;
pSampleCounts->Total_A = samples;
}