X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?p=pcsx_rearmed.git;a=blobdiff_plain;f=plugins%2Fgpu_neon%2Fgpu.c;h=02dd37210cc8e54cce8ae789606d709b9a3982f9;hp=f905f0ad8a8105e0174e632f2232a8d5c29f30ac;hb=69b09c0d33efd71ebe4886cfae41c162803683d5;hpb=1ab64c5413f79cc38f9b17520df69a4e01041046 diff --git a/plugins/gpu_neon/gpu.c b/plugins/gpu_neon/gpu.c index f905f0ad..02dd3721 100644 --- a/plugins/gpu_neon/gpu.c +++ b/plugins/gpu_neon/gpu.c @@ -8,62 +8,119 @@ * See the COPYING file in the top-level directory. */ -#include +#include #include +#include "gpu.h" #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0])) +#define unlikely(x) __builtin_expect((x), 0) +#define noinline __attribute__((noinline)) -static struct { - uint16_t vram[1024 * 512]; - uint16_t guard[1024 * 512]; // overdraw guard - uint32_t regs[16]; - union { - uint32_t reg; - struct { - uint32_t tx:4; // 0 texture page - uint32_t ty:1; - uint32_t abr:2; - uint32_t tp:2; // 7 t.p. mode (4,8,15bpp) - uint32_t dtd:1; // 9 dither - uint32_t dfe:1; - uint32_t md:1; // 11 set mask bit when drawing - uint32_t me:1; // 12 no draw on mask - uint32_t unkn:3; - uint32_t width1:1;// 16 - uint32_t width0:2; - uint32_t height:1; - uint32_t video:1; // 20 NTSC,PAL - uint32_t rgb24:1; - uint32_t inter:1; // 22 interlace on - uint32_t den:1; // 23 display not enabled - uint32_t unkn2:2; - uint32_t busy:1; // 26 !busy drawing - uint32_t img:1; // 27 ready to DMA - uint32_t com:1; // 28 ready for commands - uint32_t dma:2; // 29 off, ?, to vram, from vram - uint32_t lcf:1; // 21 odd frame/blanking? - }; - } status; - struct { - int x, y, w, h; - int y1, y2; - } screen; - uint32_t blanking; -} gpu; +#define gpu_log(fmt, ...) \ + printf("%d:%03d: " fmt, *gpu.state.frame_count, *gpu.state.hcnt, ##__VA_ARGS__) -long GPUinit(void) +//#define log_io gpu_log +#define log_io(...) +//#define log_anomaly gpu_log +#define log_anomaly(...) + +struct psx_gpu gpu __attribute__((aligned(2048))); + +static noinline void do_reset(void) { - return 0; + memset(gpu.regs, 0, sizeof(gpu.regs)); + gpu.status.reg = 0x14802000; + gpu.gp0 = 0; + gpu.regs[3] = 1; + gpu.screen.hres = gpu.screen.w = 256; + gpu.screen.vres = gpu.screen.h = 240; } -long GPUshutdown(void) +static noinline void update_width(void) { - return 0; + 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; } -uint32_t GPUreadStatus(void) +static noinline void update_height(void) +{ + int sh = gpu.screen.y2 - gpu.screen.y1; + if (gpu.status.dheight) + sh *= 2; + if (sh <= 0) + sh = gpu.screen.vres; + + gpu.screen.h = sh; +} + +static noinline void decide_frameskip(void) { - return gpu.status.reg | (gpu.blanking << 31); + if (gpu.frameskip.active) + gpu.frameskip.cnt++; + else { + gpu.frameskip.cnt = 0; + gpu.frameskip.frame_ready = 1; + } + + if (!gpu.frameskip.active && *gpu.frameskip.advice) + gpu.frameskip.active = 1; + else if (gpu.frameskip.set > 0 && gpu.frameskip.cnt < gpu.frameskip.set) + gpu.frameskip.active = 1; + else + gpu.frameskip.active = 0; +} + +static noinline void decide_frameskip_allow(uint32_t cmd_e3) +{ + // no frameskip if it decides to draw to display area, + // but not for interlace since it'll most likely always do that + uint32_t x = cmd_e3 & 0x3ff; + uint32_t y = (cmd_e3 >> 10) & 0x3ff; + gpu.frameskip.allow = 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; +} + +static noinline void get_gpu_info(uint32_t data) +{ + switch (data & 0x0f) { + 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; + break; + case 0x07: + gpu.gp0 = 2; + break; + default: + gpu.gp0 = 0; + break; + } +} + +long GPUinit(void) +{ + int ret; + ret = vout_init(); + ret |= renderer_init(); + + gpu.state.frame_count = &gpu.zero; + gpu.state.hcnt = &gpu.zero; + do_reset(); + return ret; +} + +long GPUshutdown(void) +{ + return vout_finish(); } void GPUwriteStatus(uint32_t data) @@ -72,11 +129,20 @@ void GPUwriteStatus(uint32_t data) static const short vres[4] = { 240, 480, 256, 480 }; uint32_t cmd = data >> 24; - switch (data >> 24) { + if (cmd < ARRAY_SIZE(gpu.regs)) { + if (cmd != 0 && cmd != 5 && gpu.regs[cmd] == data) + return; + gpu.regs[cmd] = data; + } + + gpu.state.fb_dirty = 1; + + switch (cmd) { case 0x00: + do_reset(); break; case 0x03: - gpu.status.den = data & 1; + gpu.status.blanking = data & 1; break; case 0x04: gpu.status.dma = data & 3; @@ -84,159 +150,391 @@ void GPUwriteStatus(uint32_t data) case 0x05: gpu.screen.x = data & 0x3ff; gpu.screen.y = (data >> 10) & 0x3ff; + if (gpu.frameskip.set) { + decide_frameskip_allow(gpu.ex_regs[3]); + if (gpu.frameskip.last_flip_frame != *gpu.state.frame_count) { + decide_frameskip(); + gpu.frameskip.last_flip_frame = *gpu.state.frame_count; + } + } + break; + case 0x06: + gpu.screen.x1 = data & 0xfff; + gpu.screen.x2 = (data >> 12) & 0xfff; + update_width(); break; case 0x07: gpu.screen.y1 = data & 0x3ff; gpu.screen.y2 = (data >> 10) & 0x3ff; + update_height(); break; case 0x08: gpu.status.reg = (gpu.status.reg & ~0x7f0000) | ((data & 0x3F) << 17) | ((data & 0x40) << 10); - gpu.screen.w = hres[(gpu.status.reg >> 16) & 7]; - gpu.screen.h = vres[(gpu.status.reg >> 19) & 3]; + gpu.screen.hres = hres[(gpu.status.reg >> 16) & 7]; + gpu.screen.vres = vres[(gpu.status.reg >> 19) & 3]; + update_width(); + update_height(); + break; + default: + if ((cmd & 0xf0) == 0x10) + get_gpu_info(data); break; } - - if (cmd < ARRAY_SIZE(gpu.regs)) - gpu.regs[cmd] = data; } -void GPUreadDataMem(uint32_t *mem, int count) +const unsigned char cmd_lengths[256] = { -} - -uint32_t GPUreadData(void) + 0, 0, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 3, 3, 3, 3, 6, 6, 6, 6, 4, 4, 4, 4, 8, 8, 8, 8, // 20 + 5, 5, 5, 5, 8, 8, 8, 8, 7, 7, 7, 7, 11, 11, 11, 11, + 2, 2, 2, 2, 0, 0, 0, 0, 3, 3, 3, 3, 3, 3, 3, 3, // 40 + 3, 3, 3, 3, 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, + 2, 2, 2, 2, 3, 3, 3, 3, 1, 1, 1, 1, 2, 2, 2, 2, // 60 + 1, 1, 1, 1, 2, 2, 2, 2, 1, 1, 1, 1, 2, 2, 2, 2, + 3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 80 + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // a0 + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // c0 + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // e0 + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 +}; + +#define VRAM_MEM_XY(x, y) &gpu.vram[(y) * 1024 + (x)] + +static inline void do_vram_line(int x, int y, uint16_t *mem, int l, int is_read) { - return 0; + uint16_t *vram = VRAM_MEM_XY(x, y); + if (is_read) + memcpy(mem, vram, l * 2); + else + memcpy(vram, mem, l * 2); } -void GPUwriteDataMem(uint32_t *mem, int count) +static int do_vram_io(uint32_t *data, int count, int is_read) { -} + int count_initial = count; + uint16_t *sdata = (uint16_t *)data; + int x = gpu.dma.x, y = gpu.dma.y; + int w = gpu.dma.w, h = gpu.dma.h; + int o = gpu.dma.offset; + int l; + count *= 2; // operate in 16bpp pixels + + if (gpu.dma.offset) { + l = w - gpu.dma.offset; + if (count < l) + l = count; + + do_vram_line(x + o, y, sdata, l, is_read); + + if (o + l < w) + o += l; + else { + o = 0; + y++; + h--; + } + sdata += l; + count -= l; + } -void GPUwriteData(uint32_t gdata) -{ + for (; h > 0 && count >= w; sdata += w, count -= w, y++, h--) { + y &= 511; + do_vram_line(x, y, sdata, w, is_read); + } + + if (h > 0 && count > 0) { + y &= 511; + do_vram_line(x, y, sdata, count, is_read); + o = count; + count = 0; + } + gpu.dma.y = y; + gpu.dma.h = h; + gpu.dma.offset = o; + + return count_initial - count / 2; } -long GPUdmaChain(uint32_t *base, uint32_t addr) +static void start_vram_transfer(uint32_t pos_word, uint32_t size_word, int is_read) { - return 0; + if (gpu.dma.h) + log_anomaly("start_vram_transfer while old unfinished\n"); + + gpu.dma.x = pos_word & 1023; + gpu.dma.y = (pos_word >> 16) & 511; + gpu.dma.w = size_word & 0xffff; // ? + gpu.dma.h = size_word >> 16; + gpu.dma.offset = 0; + + renderer_flush_queues(); + if (is_read) { + gpu.status.img = 1; + // XXX: wrong for width 1 + memcpy(&gpu.gp0, VRAM_MEM_XY(gpu.dma.x, gpu.dma.y), 4); + } + else { + renderer_invalidate_caches(gpu.dma.x, gpu.dma.y, gpu.dma.w, gpu.dma.h); + } + + log_io("start_vram_transfer %c (%d, %d) %dx%d\n", is_read ? 'r' : 'w', + gpu.dma.x, gpu.dma.y, gpu.dma.w, gpu.dma.h); } -typedef struct GPUFREEZETAG +static int check_cmd(uint32_t *data, int count) { - uint32_t ulFreezeVersion; // should be always 1 for now (set by main emu) - uint32_t ulStatus; // current gpu status - uint32_t ulControl[256]; // latest control register values - unsigned char psxVRam[1024*1024*2]; // current VRam image (full 2 MB for ZN) -} GPUFreeze_t; + int len, cmd, start, pos; + int vram_dirty = 0; -long GPUfreeze(uint32_t type, GPUFreeze_t *freeze) -{ - switch (type) { - case 1: // save - memcpy(freeze->psxVRam, gpu.vram, sizeof(gpu.vram)); - memcpy(freeze->ulControl, gpu.regs, sizeof(gpu.regs)); - freeze->ulStatus = gpu.status.reg; - freeze->ulControl[255] = gpu.blanking; // abuse free space - break; - case 0: // load - memcpy(gpu.vram, freeze->psxVRam, sizeof(gpu.vram)); - memcpy(gpu.regs, freeze->ulControl, sizeof(gpu.regs)); - gpu.status.reg = freeze->ulStatus; - gpu.blanking = freeze->ulControl[255]; - GPUwriteStatus((5 << 24) | gpu.regs[5]); - GPUwriteStatus((7 << 24) | gpu.regs[7]); - GPUwriteStatus((8 << 24) | gpu.regs[8]); + // process buffer + for (start = pos = 0; pos < count; ) + { + cmd = -1; + len = 0; + + if (gpu.dma.h) { + pos += do_vram_io(data + pos, count - pos, 0); + if (pos == count) + break; + start = pos; + } + + // do look-ahead pass to detect SR changes and VRAM i/o + while (pos < count) { + uint32_t *list = data + pos; + cmd = list[0] >> 24; + len = 1 + cmd_lengths[cmd]; + + //printf(" %3d: %02x %d\n", pos, cmd, len); + if ((cmd & 0xf4) == 0x24) { + // flat textured prim + gpu.ex_regs[1] &= ~0x1ff; + gpu.ex_regs[1] |= list[4] & 0x1ff; + } + else if ((cmd & 0xf4) == 0x34) { + // shaded textured prim + gpu.ex_regs[1] &= ~0x1ff; + gpu.ex_regs[1] |= list[5] & 0x1ff; + } + else if (cmd == 0xe3) + decide_frameskip_allow(list[0]); + + if (2 <= cmd && cmd < 0xc0) + vram_dirty = 1; + else if ((cmd & 0xf8) == 0xe0) + gpu.ex_regs[cmd & 7] = list[0]; + + if (pos + len > count) { + cmd = -1; + break; // incomplete cmd + } + if (cmd == 0xa0 || cmd == 0xc0) + break; // image i/o + pos += len; + } + + if (pos - start > 0) { + if (!gpu.frameskip.active || !gpu.frameskip.allow) + do_cmd_list(data + start, pos - start); + start = pos; + } + + if (cmd == 0xa0 || cmd == 0xc0) { + // consume vram write/read cmd + start_vram_transfer(data[pos + 1], data[pos + 2], cmd == 0xc0); + pos += len; + } + else if (cmd == -1) break; } - return 1; + gpu.status.reg &= ~0x1fff; + gpu.status.reg |= gpu.ex_regs[1] & 0x7ff; + gpu.status.reg |= (gpu.ex_regs[6] & 3) << 11; + + if (gpu.frameskip.active) + renderer_sync_ecmds(gpu.ex_regs); + gpu.state.fb_dirty |= vram_dirty; + + return count - pos; } -void GPUvBlank(int val) +void flush_cmd_buffer(void) { - gpu.blanking = !!val; + int left = check_cmd(gpu.cmd_buffer, gpu.cmd_len); + if (left > 0) + memmove(gpu.cmd_buffer, gpu.cmd_buffer + gpu.cmd_len - left, left * 4); + gpu.cmd_len = left; } -// rearmed specific - -#include "../../frontend/plugin_lib.h" -#include "../../frontend/arm_utils.h" - -static const struct rearmed_cbs *cbs; -static void *screen_buf; - -static void blit(void) +void GPUwriteDataMem(uint32_t *mem, int count) { - static uint32_t old_status, old_h; - int x = gpu.screen.x & ~3; // alignment needed by blitter - int y = gpu.screen.y; - int w = gpu.screen.w; - int h; - uint16_t *srcs; - uint8_t *dest; + int left; - srcs = &gpu.vram[y * 1024 + x]; + log_io("gpu_dma_write %p %d\n", mem, count); - h = gpu.screen.y2 - gpu.screen.y1; + if (unlikely(gpu.cmd_len > 0)) + flush_cmd_buffer(); - if (h <= 0) - return; + left = check_cmd(mem, count); + if (left) + log_anomaly("GPUwriteDataMem: discarded %d/%d words\n", left, count); +} - if ((gpu.status.reg ^ old_status) & ((7<<16)|(1<<21)) || h != old_h) // width|rgb24 change? +void GPUwriteData(uint32_t data) +{ + log_io("gpu_write %08x\n", data); + gpu.cmd_buffer[gpu.cmd_len++] = data; + if (gpu.cmd_len >= CMD_BUFFER_LEN) + flush_cmd_buffer(); +} + +long GPUdmaChain(uint32_t *rambase, uint32_t start_addr) +{ + uint32_t addr, *list; + uint32_t *llist_entry = NULL; + int len, left, count; + long cpu_cycles = 0; + + if (unlikely(gpu.cmd_len > 0)) + flush_cmd_buffer(); + + // ff7 sends it's main list twice, detect this + if (*gpu.state.frame_count == gpu.state.last_list.frame && + *gpu.state.hcnt - gpu.state.last_list.hcnt <= 1 && + gpu.state.last_list.cycles > 2048) { - old_status = gpu.status.reg; - old_h = h; - screen_buf = cbs->pl_fbdev_set_mode(w, h, gpu.status.rgb24 ? 24 : 16); + llist_entry = rambase + (gpu.state.last_list.addr & 0x1fffff) / 4; + *llist_entry |= 0x800000; } - dest = screen_buf; - if (gpu.status.rgb24) + log_io("gpu_dma_chain\n"); + addr = start_addr & 0xffffff; + for (count = 0; addr != 0xffffff; count++) { -#ifndef MAEMO - for (; h-- > 0; dest += w * 3, srcs += 1024) - { - bgr888_to_rgb888(dest, srcs, w * 3); + list = rambase + (addr & 0x1fffff) / 4; + len = list[0] >> 24; + addr = list[0] & 0xffffff; + cpu_cycles += 10; + if (len > 0) + cpu_cycles += 5 + len; + + log_io(".chain %08x #%d\n", (list - rambase) * 4, len); + + // loop detection marker + // (bit23 set causes DMA error on real machine, so + // unlikely to be ever set by the game) + list[0] |= 0x800000; + + if (len) { + left = check_cmd(list + 1, len); + if (left) + log_anomaly("GPUdmaChain: discarded %d/%d words\n", left, len); } -#else - for (; h-- > 0; dest += w * 2, srcs += 1024) - { - bgr888_to_rgb565(dest, srcs, w * 3); - } -#endif + + if (addr & 0x800000) + break; } - else - { - for (; h-- > 0; dest += w * 2, srcs += 1024) - { - bgr555_to_rgb565(dest, srcs, w * 2); - } + + // remove loop detection markers + addr = start_addr & 0x1fffff; + while (count-- > 0) { + list = rambase + addr / 4; + addr = list[0] & 0x1fffff; + list[0] &= ~0x800000; } + if (llist_entry) + *llist_entry &= ~0x800000; + + gpu.state.last_list.frame = *gpu.state.frame_count; + gpu.state.last_list.hcnt = *gpu.state.hcnt; + gpu.state.last_list.cycles = cpu_cycles; + gpu.state.last_list.addr = start_addr; - screen_buf = cbs->pl_fbdev_flip(); + return cpu_cycles; } -void GPUupdateLace(void) +void GPUreadDataMem(uint32_t *mem, int count) { - blit(); + log_io("gpu_dma_read %p %d\n", mem, count); + + if (unlikely(gpu.cmd_len > 0)) + flush_cmd_buffer(); + + if (gpu.dma.h) + do_vram_io(mem, count, 1); } -long GPUopen(void) +uint32_t GPUreadData(void) { - cbs->pl_fbdev_open(); - screen_buf = cbs->pl_fbdev_flip(); - return 0; + uint32_t ret; + + if (unlikely(gpu.cmd_len > 0)) + flush_cmd_buffer(); + + ret = gpu.gp0; + if (gpu.dma.h) + do_vram_io(&ret, 1, 1); + + log_io("gpu_read %08x\n", ret); + return ret; } -long GPUclose(void) +uint32_t GPUreadStatus(void) +{ + uint32_t ret; + + if (unlikely(gpu.cmd_len > 0)) + flush_cmd_buffer(); + + ret = gpu.status.reg; + log_io("gpu_read_status %08x\n", ret); + return ret; +} + +struct GPUFreeze +{ + uint32_t ulFreezeVersion; // should be always 1 for now (set by main emu) + uint32_t ulStatus; // current gpu status + uint32_t ulControl[256]; // latest control register values + unsigned char psxVRam[1024*1024*2]; // current VRam image (full 2 MB for ZN) +}; + +long GPUfreeze(uint32_t type, struct GPUFreeze *freeze) { - cbs->pl_fbdev_close(); - return 0; + int i; + + switch (type) { + case 1: // save + if (gpu.cmd_len > 0) + flush_cmd_buffer(); + memcpy(freeze->psxVRam, gpu.vram, sizeof(gpu.vram)); + memcpy(freeze->ulControl, gpu.regs, sizeof(gpu.regs)); + memcpy(freeze->ulControl + 0xe0, gpu.ex_regs, sizeof(gpu.ex_regs)); + freeze->ulStatus = gpu.status.reg; + break; + case 0: // load + renderer_invalidate_caches(0, 0, 1024, 512); + memcpy(gpu.vram, freeze->psxVRam, sizeof(gpu.vram)); + memcpy(gpu.regs, freeze->ulControl, sizeof(gpu.regs)); + memcpy(gpu.ex_regs, freeze->ulControl + 0xe0, sizeof(gpu.ex_regs)); + gpu.status.reg = freeze->ulStatus; + for (i = 8; i > 0; i--) { + gpu.regs[i] ^= 1; // avoid reg change detection + GPUwriteStatus((i << 24) | (gpu.regs[i] ^ 1)); + } + renderer_sync_ecmds(gpu.ex_regs); + break; + } + + return 1; } -void GPUrearmedCallbacks(const struct rearmed_cbs *cbs_) +void GPUvBlank(int is_vblank, int lcf) { - cbs = cbs_; } // vim:shiftwidth=2:expandtab