Skip to content

Commit

Permalink
gpu/bandwidth: check clipping bandwidth
Browse files Browse the repository at this point in the history
  • Loading branch information
@JaCzekanski
JaCzekanski committed Mar 23, 2021
1 parent 7adb9f0 commit 01d603f
Show file tree
Hide file tree
Showing 2 changed files with 54 additions and 98 deletions.
126 changes: 40 additions & 86 deletions gpu/bandwidth/main.c → gpu/bandwidth/main.cpp
View file
Original file line number Diff line number Diff line change
@@ -1,16 +1,17 @@
#include <stdio.h>
#include <stdlib.h>
#include <psxapi.h>
#include <psxgpu.h>
#include <psxetc.h>
#include <io.h>
#include <gpu.h>
#include <timer.h>

typedef char bool;

#define SCR_W 320
#define SCR_H 240

bool semitransparent = false;

uint32_t getTimer() {
uint32_t value = readTimer(1);
if (timerDidOverflow(1)) value += 0xffff;
Expand All @@ -28,7 +29,7 @@ void fillScreen() {
DrawPrim(&f);
}

void rectScreen(bool semitransparent) {
void rectScreen() {
TILE t;
setTile(&t);
setSemiTrans(&t, semitransparent);
Expand All @@ -39,7 +40,7 @@ void rectScreen(bool semitransparent) {
DrawPrim(&t);
}

void rectTexturedScreen(bool semitransparent) {
void rectTexturedScreen() {
SPRT s;
setSprt(&s);
setSemiTrans(&s, semitransparent);
Expand All @@ -54,15 +55,17 @@ void rectTexturedScreen(bool semitransparent) {
DrawPrim(&s);
}

void quadScreen(bool semitransparent) {
int xOffset = 0;

void quadScreen() {
POLY_F4 p;
setPolyF4(&p);
setSemiTrans(&p, semitransparent);
setXY4(&p,
0, 0,
SCR_W, 0,
0, SCR_H,
SCR_W, SCR_H
xOffset + 0, 0,
xOffset + SCR_W, 0,
xOffset + 0, SCR_H,
xOffset + SCR_W, SCR_H
);
setRGB0(&p, rand()%255, rand()%255, rand()%255);

Expand All @@ -75,7 +78,7 @@ void quadScreen(bool semitransparent) {
(p)->u2 = _u0, (p)->v2 = _v0+(_h), \
(p)->u3 = _u0+(_w), (p)->v3 = _v0+(_h)

void quadTexturedScreen(bool semitransparent) {
void quadTexturedScreen() {
POLY_FT4 p;
setPolyFT4(&p);
setSemiTrans(&p, semitransparent);
Expand All @@ -97,7 +100,7 @@ void quadTexturedScreen(bool semitransparent) {
DrawPrim(&p);
}

const int callCount = 400;
constexpr int callCount = 400;

void calculate(const char* test, uint16_t hblanks) {
const int scanlinesPerFrame = 263;
Expand All @@ -118,96 +121,47 @@ void calculate(const char* test, uint16_t hblanks) {
printf("%-30s dT: %5d.%-5d ms (hblanks: %5d), speed: %d MB/s\n", test, dt_q, dt_r, diff, megaBytesPerSecond);
}

const char* testCases[9] = {
"FillScreen GP0(2)",
"Rectangle",
"Rectangle (semitransparent)",
"Rectangle textured",
"Rectangle textured (semi)",
"Polygon quad",
"Polygon quad (semi)",
"Polygon quad textured",
"Polygon quad textured (semi)"
};

uint16_t* buffer = (uint16_t*)0x801a0000;

void testVramToCpu() {
DrawSync(0);
resetTimer(1);
for (int i = 0; i < callCount; i++) {
vramReadDMA(0, 0, SCR_W, SCR_H, buffer);
}
calculate("vramToCpu", getTimer());
}

void testCpuToVram() {
DrawSync(0);
resetTimer(1);
for (int i = 0; i < callCount; i++) {
vramWriteDMA(0, 0, SCR_W, SCR_H, buffer);
}
calculate("cpuToVram", getTimer());
}

void testVramToVram() {
DrawSync(0);
resetTimer(1);
for (int i = 0; i < callCount; i++) {
vramToVramCopy(0, 0, 320, 0, SCR_W, SCR_H);
}
calculate("vramToVram", getTimer());
}

int main() {
initVideo(SCR_W, SCR_H);
SetVideoMode(MODE_NTSC);
printf("\ngpu/bandwidth\n");

uint16_t oldTimer1Mode = initTimer(1, 1); // Timer1, HBlank

testVramToCpu();
testCpuToVram();
testVramToVram();

for (int test = 0; test<9; test++) {
auto test = [](const char* name, void (*func)(), bool semi = false) {
clearScreen();

DrawSync(0);
resetTimer(1);
semitransparent = semi;
for (int i = 0; i < callCount; i++) {
switch (test) {
case 0:
fillScreen();
break;
case 1:
rectScreen(false);
break;
case 2:
rectScreen(true);
break;
case 3:
rectTexturedScreen(false);
break;
case 4:
rectTexturedScreen(true);
break;
case 5:
quadScreen(false);
break;
case 6:
quadScreen(true);
break;
case 7:
quadTexturedScreen(false);
break;
case 8:
quadTexturedScreen(true);
break;
}
func();
}
calculate(testCases[test], getTimer());
}
calculate(name, getTimer());
};

EnterCriticalSection();
test("vramToCpu", []() { vramReadDMA(0, 0, SCR_W, SCR_H, buffer); });
test("cpuToVram", []() { vramWriteDMA(0, 0, SCR_W, SCR_H, buffer); });
test("vramToVram", []() { vramToVramCopy(0, 0, 320, 0, SCR_W, SCR_H); });
test("FillScreen GP0(2)", fillScreen);
test("Rectangle", rectScreen, false);
test("Rectangle (semitransparent)", rectScreen, true);
test("Rectangle textured", rectTexturedScreen, false);
test("Rectangle textured (semi)", rectTexturedScreen, true);
test("Polygon quad", quadScreen, false);
test("Polygon quad (semi)", quadScreen, true);
test("Polygon quad textured", quadTexturedScreen, false);
test("Polygon quad textured (semi)", quadTexturedScreen, true);

xOffset = -SCR_W/4;
test("Polygon quad (1/4 off screen)", quadScreen, false);

xOffset = -SCR_W/2;
test("Polygon quad (1/2 off screen)", quadScreen, false);
ExitCriticalSection();

restoreTimer(1, oldTimer1Mode);

Expand Down
26 changes: 14 additions & 12 deletions gpu/bandwidth/psx.log
View file
Original file line number Diff line number Diff line change
@@ -1,14 +1,16 @@
gpu/bandwidth
vramToCpu dT: 987.8140 ms (hblanks: 15583), speed: 60 MB/s
cpuToVram dT: 766.4520 ms (hblanks: 12092), speed: 78 MB/s
vramToVram dT: 1211.420 ms (hblanks: 19110), speed: 49 MB/s
FillScreen GP0(2) dT: 76.4720 ms (hblanks: 1204), speed: 789 MB/s
Rectangle dT: 476.720 ms (hblanks: 7512), speed: 126 MB/s
Rectangle (semitransparent) dT: 717.3740 ms (hblanks: 11318), speed: 83 MB/s
Rectangle textured dT: 951.5220 ms (hblanks: 15012), speed: 63 MB/s
Rectangle textured (semi) dT: 951.5220 ms (hblanks: 15012), speed: 63 MB/s
Polygon quad dT: 483.15260 ms (hblanks: 7637), speed: 124 MB/s
Polygon quad (semi) dT: 729.4380 ms (hblanks: 11508), speed: 82 MB/s
Polygon quad textured dT: 2501.2220 ms (hblanks: 39468), speed: 23 MB/s
Polygon quad textured (semi) dT: 2501.3220 ms (hblanks: 39469), speed: 23 MB/s
vramToCpu dT: 999.5780 ms (hblanks: 15770), speed: 60 MB/s
cpuToVram dT: 772.12840 ms (hblanks: 12195), speed: 77 MB/s
vramToVram dT: 1212.6640 ms (hblanks: 19132), speed: 49 MB/s
FillScreen GP0(2) dT: 77.940 ms (hblanks: 1216), speed: 779 MB/s
Rectangle dT: 481.12820 ms (hblanks: 7603), speed: 124 MB/s
Rectangle (semitransparent) dT: 725.14500 ms (hblanks: 11455), speed: 82 MB/s
Rectangle textured dT: 959.4980 ms (hblanks: 15138), speed: 62 MB/s
Rectangle textured (semi) dT: 959.5980 ms (hblanks: 15139), speed: 62 MB/s
Polygon quad dT: 489.10580 ms (hblanks: 7727), speed: 122 MB/s
Polygon quad (semi) dT: 738.2360 ms (hblanks: 11648), speed: 81 MB/s
Polygon quad textured dT: 2530.14600 ms (hblanks: 39938), speed: 23 MB/s
Polygon quad textured (semi) dT: 2530.14600 ms (hblanks: 39938), speed: 23 MB/s
Polygon quad (1/4 off screen) dT: 370.6400 ms (hblanks: 5845), speed: 162 MB/s
Polygon quad (1/2 off screen) dT: 251.7220 ms (hblanks: 3968), speed: 239 MB/s
Done

0 comments on commit 01d603f

Please sign in to comment.