mirror/psxrecomp
1
0
Fork 0
mirror of https://github.com/RetroPortingToolKit/psxrecomp synced 2026-10-26 02:11:36 -04:00
Code Issues Packages Projects Releases Wiki Activity

Restore depth-pass GL state without synchronous driver queries

Browse Source
This commit is contained in:
Shokunin
2026-10-21 09:01:38 -07:00
parent c5639b0788
commit 1ef10b7f79
3 changed files with 53 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/internal/FAITHFUL_TIMING_PLAN.md
+2
View File
@@ -213,6 +213,8 @@ on a fixed region -> next.
## 5. Status / Log (update every session)
- **2026-10-21 (BRIEF-4 depth state):** Removed per-batch GL state round trips, including the mask-check exception absent in the old experiment. Real-context state restoration and image/readback controls pass on Windows and Mac. Fullscreen quiet route remains unaccepted (8 completion gaps, 53.2% p99 busy). Investigating GPU query spans that include host pacing and provoke expensive dynamic-resolution reallocations.
- **2026-10-21 (BRIEF-4 Stage 1):** Reused #722 display-height cap and added opt-in `supersample_disable_at_lines`; unchanged defaults for other titles. Resolution, replay, config-parser and R4 default checks pass on Windows and macOS. R4 fullscreen 8K route cap restores 60 Hz but completion gaps remain (6 quiet / 7 loaded); no stage acceptance claimed. See R4 `analysis/STATUS.md`.
- **2026-10-21 (BRIEF-4 Stage 1):** Port recorded-route measurement support from `a9da7fce`; validation pending. No guest timing changes.
runtime/src/gpu_gl_renderer.c
+23 -10
View File
@@ -2419,11 +2419,24 @@ static void depth_apply_ex(int mode, int tex) {
}
/* After the colour pass of mode 1: the depth-only pass (n vertices of the
* bound buffer), then back to the default state. */
static void depth_restore_ex(int mode, int tex, int n) {
static int depth_state_fast(void) {
static int enabled = -1;
if (enabled < 0) {
const char *e = getenv("PSX_GL_STATE_QUERY_FAST");
enabled = !(e && e[0] == '0');
}
return enabled;
}
static void depth_restore_ex(int mode, int tex, int n, int semi, int mask, int check) {
if (mode == 0) return;
if (mode == 1) {
p_glUniform1f(tex ? s_tex_uZbias : s_geo_uZbias, 0.0f);
GLboolean st = glIsEnabled(GL_STENCIL_TEST), bl = glIsEnabled(GL_BLEND);
/* The preceding colour pass establishes these states. Avoid two
* synchronous queries to the threaded driver for every depth batch. */
GLboolean st = !(tex && semi == 4 && !mask && !check), bl = semi >= 0;
if (!depth_state_fast()) {
st = glIsEnabled(GL_STENCIL_TEST); bl = glIsEnabled(GL_BLEND);
}
glDisable(GL_STENCIL_TEST); glDisable(GL_BLEND);
glColorMask(GL_FALSE, GL_FALSE, GL_FALSE, GL_FALSE);
glDepthMask(GL_TRUE);
@@ -3999,7 +4012,7 @@ static void hiw_replay_wide(void) {
s_hq_v + c->vfirst, PSXGL_STREAM_DRAW);
depth_apply_ex(c->depth, 1);
tex_draw_passes_ex(c->vcount, c->semi, c->mask, c->check, 0, -1);
depth_restore_ex(c->depth, 1, c->vcount);
depth_restore_ex(c->depth, 1, c->vcount, c->semi, c->mask, c->check);
} else {
if (cur != HQ_GEO) {
p_glUseProgram(s_geo_prog);
@@ -4016,7 +4029,7 @@ static void hiw_replay_wide(void) {
s_hq_v + c->vfirst, PSXGL_STREAM_DRAW);
depth_apply_ex(c->depth, 0);
BCLS_DRAW(c->semi, c->depth, c->mask, c->check, -1, glDrawArrays(GL_TRIANGLES, 0, c->vcount));
depth_restore_ex(c->depth, 0, c->vcount);
depth_restore_ex(c->depth, 0, c->vcount, c->semi, c->mask, c->check);
}
}
if (!any) return;
@@ -4083,7 +4096,7 @@ static void hiw_flush_queue(void) {
s_hq_v + c->vfirst, PSXGL_STREAM_DRAW);
depth_apply_ex(c->depth, 1);
tex_draw_passes_ex(c->vcount, c->semi, c->mask, c->check, 0, -1);
depth_restore_ex(c->depth, 1, c->vcount);
depth_restore_ex(c->depth, 1, c->vcount, c->semi, c->mask, c->check);
} else {
if (cur != HQ_GEO) {
p_glUseProgram(s_geo_prog);
@@ -4097,7 +4110,7 @@ static void hiw_flush_queue(void) {
s_hq_v + c->vfirst, PSXGL_STREAM_DRAW);
depth_apply_ex(c->depth, 0);
BCLS_DRAW(c->semi, c->depth, c->mask, c->check, -1, glDrawArrays(GL_TRIANGLES, 0, c->vcount));
depth_restore_ex(c->depth, 0, c->vcount);
depth_restore_ex(c->depth, 0, c->vcount, c->semi, c->mask, c->check);
}
}
}
@@ -4144,7 +4157,7 @@ static void flush_tex_batch_impl(void) {
depth_apply_ex(dmode, 1);
tex_batch_draw_passes(nverts, semi);
depth_restore_ex(dmode, 1, nverts);
depth_restore_ex(dmode, 1, nverts, semi, s_tb_mask, s_mask_check);
/* Native-wide mirror — skipped for a batch fully inside the 4:3 frame (its
* centre content comes from the present-time canonical blit; nothing to add
@@ -4173,7 +4186,7 @@ static void flush_tex_batch_impl(void) {
wide_set_bd_scale(s_tex_uXscale, s_tex_uXcenter);
depth_apply_ex(dmode, 1);
if (s_ws_ablate != 2) tex_batch_draw_passes(nverts, semi);
depth_restore_ex(dmode, 1, nverts);
depth_restore_ex(dmode, 1, nverts, semi, s_tb_mask, s_mask_check);
wide_clear_bd_scale(s_tex_uXscale, s_tex_uXcenter);
wide_target_end(s_tex_uXoff, s_tex_uXhalf);
gl_perf_mirror_end();
@@ -4285,7 +4298,7 @@ static void flush_flat_batch_impl(void) {
depth_apply_ex(dmode, 0);
BCLS_DRAW(semi, dmode, mask, s_mask_check, b3d, if (nl) flat_batch_draw_hr_lines(nverts, nl);
else glDrawArrays(fmode, 0, nverts));
depth_restore_ex(dmode, 0, nverts);
depth_restore_ex(dmode, 0, nverts, semi, mask, s_mask_check);
int mirror = g_wide_cur && !s_wide_suppress && s_ws_ablate != 1 &&
!(!g_ws_bd_stretch_on && mirror_flat_batch_center_only(nverts));
if (mirror) {
@@ -4306,7 +4319,7 @@ static void flush_flat_batch_impl(void) {
wide_set_bd_scale(s_geo_uXscale, s_geo_uXcenter);
depth_apply_ex(dmode, 0);
if (s_ws_ablate != 2) BCLS_DRAW(semi, dmode, mask, s_mask_check, b3d, glDrawArrays(fmode, 0, nverts));
depth_restore_ex(dmode, 0, nverts);
depth_restore_ex(dmode, 0, nverts, semi, mask, s_mask_check);
wide_clear_bd_scale(s_geo_uXscale, s_geo_uXcenter);
wide_target_end(s_geo_uXoff, s_geo_uXhalf);
gl_perf_mirror_end();
runtime/tests/test_gl_render_thread.c
+28
View File
@@ -85,6 +85,11 @@ void host_osd_present_done(void){}
int psx_rewind_overlay_image(const uint32_t **p,int *w,int *h){*p=NULL;*w=*h=0;return 0;}
float psx_rewind_slide(void){return 0.0f;}
int psx_savestate_menu_overlay_image(const uint32_t **p,int *w,int *h){*p=NULL;*w=*h=0;return 0;}
/* No frontend menu or telemetry consumer in this renderer-only fixture. */
int psx_host_menu_overlay_image(const uint32_t **p,int *w,int *h){*p=NULL;*w=*h=0;return 0;}
void psx_video_telemetry_compose(int r,int s,int w,int h,int dw,int dh,int vw,int vh){
(void)r;(void)s;(void)w;(void)h;(void)dw;(void)dh;(void)vw;(void)vh;}
void psx_video_telemetry_present(int result){(void)result;}
int present_shot_take(char *out,int n){(void)out;(void)n;return 0;}
void present_shot_done(int ok){(void)ok;}
void psx_host_sleep_ms(uint32_t ms){(void)ms;}
@@ -272,6 +277,29 @@ int main(int argc, char **argv) {
gl_renderer_set_swap_interval(0);
if (!gl_renderer_init_context(win)) { fprintf(stderr, "FAIL context\n"); return 2; }
check(gr_backend() == GR_BACKEND_OPENGL, "OpenGL backend");
/* Compare restoration with real GL state after every colour-pass branch,
* including mask-check, dual-source blending and bloom marking. */
for (int bloom = 0; bloom <= 1; ++bloom)
for (int tex = 0; tex <= 1; ++tex)
for (int semi = -1; semi <= (tex ? 4 : 3); ++semi)
for (int mask = 0; mask <= 1; ++mask)
for (int check_mask = 0; check_mask <= 1; ++check_mask) {
s_bcls_on = bloom;
p_glUseProgram(tex ? s_tex_prog : s_geo_prog);
p_glBindVertexArray(tex ? s_tex_vao : s_geo_vao);
depth_apply_ex(1, tex);
if (tex) tex_draw_passes_ex(0, semi, mask, check_mask, 0, -1);
else {
if (semi >= 0) apply_psx_blend(semi); else glDisable(GL_BLEND);
mask_stencil_ex(mask, check_mask);
BCLS_DRAW(semi, 1, mask, check_mask, -1, glDrawArrays(GL_TRIANGLES, 0, 0));
}
GLboolean stencil = glIsEnabled(GL_STENCIL_TEST), blend = glIsEnabled(GL_BLEND);
depth_restore_ex(1, tex, 0, semi, mask, check_mask);
check(glIsEnabled(GL_STENCIL_TEST) == stencil, "depth pass restores stencil");
check(glIsEnabled(GL_BLEND) == blend, "depth pass restores blend");
}
s_bcls_on = 0;
textures();
gr_wide_configure(426, 53);
if (threaded) {