*/
#include <stdio.h>
+#include <stdlib.h>
#include <string.h>
#include "gpu.h"
#define noinline
#endif
-#define gpu_log(fmt, ...) \
- printf("%d:%03d: " fmt, *gpu.state.frame_count, *gpu.state.hcnt, ##__VA_ARGS__)
-
//#define log_io gpu_log
#define log_io(...)
-//#define log_anomaly gpu_log
-#define log_anomaly(...)
struct psx_gpu gpu;
gpu.regs[3] = 1;
gpu.screen.hres = gpu.screen.w = 256;
gpu.screen.vres = gpu.screen.h = 240;
+ gpu.screen.x = gpu.screen.y = 0;
+ renderer_notify_res_change();
}
static noinline void update_width(void)
{
+ static const short hres_all[8] = { 256, 368, 320, 368, 512, 368, 640, 368 };
+ static const uint8_t hdivs[8] = { 10, 7, 8, 7, 5, 7, 4, 7 };
+ uint8_t hdiv = hdivs[(gpu.status >> 16) & 7];
+ int hres = hres_all[(gpu.status >> 16) & 7];
+ int pal = gpu.status & PSX_GPU_STATUS_PAL;
int sw = gpu.screen.x2 - gpu.screen.x1;
- if (sw <= 0 || sw >= 2560)
- // full width
- gpu.screen.w = gpu.screen.hres;
- else
- gpu.screen.w = sw * gpu.screen.hres / 2560;
+ int x = 0, x_auto;
+ if (sw <= 0)
+ /* nothing displayed? */;
+ else {
+ int s = pal ? 656 : 608; // or 600? pal is just a guess
+ x = (gpu.screen.x1 - s) / hdiv;
+ x = (x + 1) & ~1; // blitter limitation
+ sw /= hdiv;
+ sw = (sw + 2) & ~3; // according to nocash
+ switch (gpu.state.screen_centering_type) {
+ case 1:
+ break;
+ case 2:
+ x = gpu.state.screen_centering_x;
+ break;
+ default:
+ // correct if slightly miscentered
+ x_auto = (hres - sw) / 2 & ~3;
+ if ((uint32_t)x_auto <= 8u && abs(x) < 24)
+ x = x_auto;
+ }
+ if (x + sw > hres)
+ sw = hres - x;
+ // .x range check is done in vout_update()
+ }
+ // reduce the unpleasant right border that a few games have
+ if (gpu.state.screen_centering_type == 0
+ && x <= 4 && hres - (x + sw) >= 4)
+ hres -= 4;
+ gpu.screen.x = x;
+ gpu.screen.w = sw;
+ gpu.screen.hres = hres;
+ gpu.state.dims_changed = 1;
+ //printf("xx %d %d -> %2d, %d / %d\n",
+ // gpu.screen.x1, gpu.screen.x2, x, sw, hres);
}
static noinline void update_height(void)
{
- // TODO: emulate this properly..
+ int pal = gpu.status & PSX_GPU_STATUS_PAL;
+ int dheight = gpu.status & PSX_GPU_STATUS_DHEIGHT;
+ int y = gpu.screen.y1 - (pal ? 39 : 16); // 39 for spyro
int sh = gpu.screen.y2 - gpu.screen.y1;
- if (gpu.status & PSX_GPU_STATUS_DHEIGHT)
- sh *= 2;
- if (sh <= 0 || sh > gpu.screen.vres)
- sh = gpu.screen.vres;
-
+ int center_tol = 16;
+ int vres = 240;
+
+ if (pal && (sh > 240 || gpu.screen.vres == 256))
+ vres = 256;
+ if (dheight)
+ y *= 2, sh *= 2, vres *= 2, center_tol *= 2;
+ if (sh <= 0)
+ /* nothing displayed? */;
+ else {
+ switch (gpu.state.screen_centering_type) {
+ case 1:
+ break;
+ case 2:
+ y = gpu.state.screen_centering_y;
+ break;
+ default:
+ // correct if slightly miscentered
+ if ((uint32_t)(vres - sh) <= 1 && abs(y) <= center_tol)
+ y = 0;
+ }
+ if (y + sh > vres)
+ sh = vres - y;
+ }
+ gpu.screen.y = y;
gpu.screen.h = sh;
+ gpu.screen.vres = vres;
+ gpu.state.dims_changed = 1;
+ //printf("yy %d %d -> %d, %d / %d\n",
+ // gpu.screen.y1, gpu.screen.y2, y, sh, vres);
}
static noinline void decide_frameskip(void)
uint32_t x = cmd_e3 & 0x3ff;
uint32_t y = (cmd_e3 >> 10) & 0x3ff;
gpu.frameskip.allow = (gpu.status & PSX_GPU_STATUS_INTERLACE) ||
- (uint32_t)(x - gpu.screen.x) >= (uint32_t)gpu.screen.w ||
- (uint32_t)(y - gpu.screen.y) >= (uint32_t)gpu.screen.h;
+ (uint32_t)(x - gpu.screen.src_x) >= (uint32_t)gpu.screen.w ||
+ (uint32_t)(y - gpu.screen.src_y) >= (uint32_t)gpu.screen.h;
return gpu.frameskip.allow;
}
case 0x02:
case 0x03:
case 0x04:
- case 0x05:
gpu.gp0 = gpu.ex_regs[data & 7] & 0xfffff;
break;
- case 0x06:
- gpu.gp0 = gpu.ex_regs[5] & 0xfffff;
+ case 0x05:
+ gpu.gp0 = gpu.ex_regs[5] & 0x3fffff;
break;
case 0x07:
gpu.gp0 = 2;
break;
default:
- gpu.gp0 = 0;
+ // gpu.gp0 unchanged
break;
}
}
ret = vout_init();
ret |= renderer_init();
+ memset(&gpu.state, 0, sizeof(gpu.state));
+ memset(&gpu.frameskip, 0, sizeof(gpu.frameskip));
+ gpu.zero = 0;
gpu.state.frame_count = &gpu.zero;
gpu.state.hcnt = &gpu.zero;
- gpu.frameskip.active = 0;
gpu.cmd_len = 0;
do_reset();
void GPUwriteStatus(uint32_t data)
{
- static const short hres[8] = { 256, 368, 320, 384, 512, 512, 640, 640 };
- static const short vres[4] = { 240, 480, 256, 480 };
uint32_t cmd = data >> 24;
if (cmd < ARRAY_SIZE(gpu.regs)) {
do_cmd_reset();
break;
case 0x03:
- if (data & 1)
+ if (data & 1) {
gpu.status |= PSX_GPU_STATUS_BLANKING;
+ gpu.state.dims_changed = 1; // for hud clearing
+ }
else
gpu.status &= ~PSX_GPU_STATUS_BLANKING;
break;
gpu.status |= PSX_GPU_STATUS_DMA(data & 3);
break;
case 0x05:
- gpu.screen.x = data & 0x3ff;
- gpu.screen.y = (data >> 10) & 0x1ff;
+ gpu.screen.src_x = data & 0x3ff;
+ gpu.screen.src_y = (data >> 10) & 0x1ff;
+ renderer_notify_scanout_x_change(gpu.screen.src_x, gpu.screen.hres);
if (gpu.frameskip.set) {
decide_frameskip_allow(gpu.ex_regs[3]);
if (gpu.frameskip.last_flip_frame != *gpu.state.frame_count) {
break;
case 0x08:
gpu.status = (gpu.status & ~0x7f0000) | ((data & 0x3F) << 17) | ((data & 0x40) << 10);
- gpu.screen.hres = hres[(gpu.status >> 16) & 7];
- gpu.screen.vres = vres[(gpu.status >> 19) & 3];
update_width();
update_height();
renderer_notify_res_change();
if (is_read) {
gpu.status |= PSX_GPU_STATUS_IMG;
// XXX: wrong for width 1
- memcpy(&gpu.gp0, VRAM_MEM_XY(gpu.dma.x, gpu.dma.y), 4);
+ gpu.gp0 = LE32TOH(*(uint32_t *) VRAM_MEM_XY(gpu.dma.x, gpu.dma.y));
gpu.state.last_vram_read_frame = *gpu.state.frame_count;
}
gpu.status &= ~PSX_GPU_STATUS_IMG;
else
renderer_update_caches(gpu.dma_start.x, gpu.dma_start.y,
- gpu.dma_start.w, gpu.dma_start.h);
+ gpu.dma_start.w, gpu.dma_start.h, 0);
}
static noinline int do_cmd_list_skip(uint32_t *data, int count, int *last_cmd)
flush_cmd_buffer();
}
-long GPUdmaChain(uint32_t *rambase, uint32_t start_addr)
+long GPUdmaChain(uint32_t *rambase, uint32_t start_addr, uint32_t *progress_addr)
{
uint32_t addr, *list, ld_addr = 0;
int len, left, count;
if (len > 0)
cpu_cycles += 5 + len;
- log_io(".chain %08x #%d\n", (list - rambase) * 4, len);
+ log_io(".chain %08lx #%d+%d\n",
+ (long)(list - rambase) * 4, len, gpu.cmd_len);
+ if (unlikely(gpu.cmd_len > 0)) {
+ if (gpu.cmd_len + len > ARRAY_SIZE(gpu.cmd_buffer)) {
+ log_anomaly("cmd_buffer overflow, likely garbage commands\n");
+ gpu.cmd_len = 0;
+ }
+ memcpy(gpu.cmd_buffer + gpu.cmd_len, list + 1, len * 4);
+ gpu.cmd_len += len;
+ flush_cmd_buffer();
+ continue;
+ }
if (len) {
left = do_cmd_buffer(list + 1, len);
- if (left)
- log_anomaly("GPUdmaChain: discarded %d/%d words\n", left, len);
+ if (left) {
+ memcpy(gpu.cmd_buffer, list + 1 + len - left, left * 4);
+ gpu.cmd_len = left;
+ log_anomaly("GPUdmaChain: %d/%d words left\n", left, len);
+ }
}
+ if (progress_addr) {
+ *progress_addr = addr;
+ break;
+ }
#define LD_THRESHOLD (8*1024)
if (count >= LD_THRESHOLD) {
if (count == LD_THRESHOLD) {
flush_cmd_buffer();
ret = gpu.gp0;
- if (gpu.dma.h)
+ if (gpu.dma.h) {
+ ret = HTOLE32(ret);
do_vram_io(&ret, 1, 1);
+ ret = LE32TOH(ret);
+ }
log_io("gpu_read %08x\n", ret);
return ret;
GPUwriteStatus((i << 24) | (gpu.regs[i] ^ 1));
}
renderer_sync_ecmds(gpu.ex_regs);
- renderer_update_caches(0, 0, 1024, 512);
+ renderer_update_caches(0, 0, 1024, 512, 1);
break;
}
}
vout_update();
+ if (gpu.state.enhancement_active && !gpu.state.enhancement_was_active)
+ renderer_update_caches(0, 0, 1024, 512, 1);
+ gpu.state.enhancement_was_active = gpu.state.enhancement_active;
gpu.state.fb_dirty = 0;
gpu.state.blanked = 0;
}
gpu.state.frame_count = cbs->gpu_frame_count;
gpu.state.allow_interlace = cbs->gpu_neon.allow_interlace;
gpu.state.enhancement_enable = cbs->gpu_neon.enhancement_enable;
+ if (gpu.state.screen_centering_type != cbs->screen_centering_type
+ || gpu.state.screen_centering_x != cbs->screen_centering_x
+ || gpu.state.screen_centering_y != cbs->screen_centering_y) {
+ gpu.state.screen_centering_type = cbs->screen_centering_type;
+ gpu.state.screen_centering_x = cbs->screen_centering_x;
+ gpu.state.screen_centering_y = cbs->screen_centering_y;
+ update_width();
+ update_height();
+ }
gpu.mmap = cbs->mmap;
gpu.munmap = cbs->munmap;