src/gallium/drivers/r300/r300_surface.c

   1 /*
   2  * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
   3  *                Joakim Sindholt <opensource@zhasha.com>
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the "Software"),
   7  * to deal in the Software without restriction, including without limitation
   8  * on the rights to use, copy, modify, merge, publish, distribute, sub
   9  * license, and/or sell copies of the Software, and to permit persons to whom
  10  * the Software is furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice (including the next
  13  * paragraph) shall be included in all copies or substantial portions of the
  14  * Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  20  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  21  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  22  * USE OR OTHER DEALINGS IN THE SOFTWARE. */
  23
  24 #include "r300_surface.h"
  25
  26 static void r300_surface_setup(struct r300_context* r300,
  27                                struct r300_texture* dest,
  28                                unsigned x, unsigned y,
  29                                unsigned w, unsigned h)
  30 {
  31     struct r300_capabilities* caps = r300_screen(r300->context.screen)->caps;
  32     unsigned pixpitch = dest->stride / dest->tex.block.size;
  33     CS_LOCALS(r300);
  34
  35     r300_emit_blend_state(r300, &blend_clear_state);
  36     r300_emit_blend_color_state(r300, &blend_color_clear_state);
  37     r300_emit_dsa_state(r300, &dsa_clear_state);
  38     r300_emit_rs_state(r300, &rs_clear_state);
  39
  40     BEGIN_CS(24);
  41
  42     /* Viewport setup */
  43     OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE, 6);
  44     OUT_CS_32F((float)w);
  45     OUT_CS_32F((float)x);
  46     OUT_CS_32F((float)h);
  47     OUT_CS_32F((float)y);
  48     OUT_CS_32F(1.0);
  49     OUT_CS_32F(0.0);
  50
  51     OUT_CS_REG(R300_VAP_VTE_CNTL, R300_VPORT_X_SCALE_ENA |
  52             R300_VPORT_X_OFFSET_ENA |
  53             R300_VPORT_Y_SCALE_ENA |
  54             R300_VPORT_Y_OFFSET_ENA |
  55             R300_VTX_XY_FMT | R300_VTX_Z_FMT);
  56
  57     /* Pixel scissors. */
  58     OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL, 2);
  59     if (caps->is_r500) {
  60         OUT_CS((x << R300_SCISSORS_X_SHIFT) | (y << R300_SCISSORS_Y_SHIFT));
  61         OUT_CS((w << R300_SCISSORS_X_SHIFT) | (h << R300_SCISSORS_Y_SHIFT));
  62     } else {
  63         /* Non-R500 chipsets have an offset of 1440 in their scissors. */
  64         OUT_CS(((x + 1440) << R300_SCISSORS_X_SHIFT) |
  65                 ((y + 1440) << R300_SCISSORS_Y_SHIFT));
  66         OUT_CS(((w + 1440) << R300_SCISSORS_X_SHIFT) |
  67                 ((h + 1440) << R300_SCISSORS_Y_SHIFT));
  68     }
  69
  70     /* Flush colorbuffer and blend caches. */
  71     OUT_CS_REG(R300_RB3D_DSTCACHE_CTLSTAT,
  72         R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D |
  73         R300_RB3D_DSTCACHE_CTLSTAT_DC_FINISH_SIGNAL);
  74     OUT_CS_REG(R300_ZB_ZCACHE_CTLSTAT,
  75         R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE |
  76         R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
  77
  78     /* Setup colorbuffer. */
  79     OUT_CS_REG_SEQ(R300_RB3D_COLOROFFSET0, 1);
  80     OUT_CS_RELOC(dest->buffer, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
  81     OUT_CS_REG(R300_RB3D_COLORPITCH0, pixpitch |
  82         r300_translate_colorformat(dest->tex.format));
  83     OUT_CS_REG(RB3D_COLOR_CHANNEL_MASK, 0xf);
  84
  85     END_CS;
  86 }
  87
  88 /* Provides pipe_context's "surface_fill". Commonly used for clearing
  89  * buffers. */
  90 static void r300_surface_fill(struct pipe_context* pipe,
  91                               struct pipe_surface* dest,
  92                               unsigned x, unsigned y,
  93                               unsigned w, unsigned h,
  94                               unsigned color)
  95 {
  96     int i;
  97     float r, g, b, a, depth;
  98     struct r300_context* r300 = r300_context(pipe);
  99     struct r300_capabilities* caps = r300_screen(pipe->screen)->caps;
 100     struct r300_texture* tex = (struct r300_texture*)dest->texture;
 101     unsigned pixpitch = tex->stride / tex->tex.block.size;
 102     boolean invalid = FALSE;
 103     CS_LOCALS(r300);
 104
 105     a = (float)((color >> 24) & 0xff) / 255.0f;
 106     r = (float)((color >> 16) & 0xff) / 255.0f;
 107     g = (float)((color >>  8) & 0xff) / 255.0f;
 108     b = (float)((color >>  0) & 0xff) / 255.0f;
 109     debug_printf("r300: Filling surface %p at (%d,%d),"
 110         " dimensions %dx%d (pixel pitch %d), color 0x%x\n",
 111         dest, x, y, w, h, pixpitch, color);
 112
 113     /* Fallback? */
 114     if (FALSE) {
 115 fallback:
 116         debug_printf("r300: Falling back on surface clear...");
 117         util_surface_fill(pipe, dest, x, y, w, h, color);
 118         return;
 119     }
 120
 121     /* Make sure our target BO is okay. */
 122 validate:
 123     if (!r300->winsys->add_buffer(r300->winsys, tex->buffer,
 124                 0, RADEON_GEM_DOMAIN_VRAM)) {
 125         r300->context.flush(&r300->context, 0, NULL);
 126         goto validate;
 127     }
 128     if (!r300->winsys->validate(r300->winsys)) {
 129         r300->context.flush(&r300->context, 0, NULL);
 130         if (invalid) {
 131             debug_printf("r300: Stuck in validation loop, gonna fallback.");
 132             goto fallback;
 133         }
 134         invalid = TRUE;
 135         goto validate;
 136     }
 137
 138     r300_surface_setup(r300, tex, x, y, w, h);
 139
 140     /* Vertex shader setup */
 141     if (caps->has_tcl) {
 142         r300_emit_vertex_program_code(r300, &r300_passthrough_vertex_shader, 0);
 143     } else {
 144         BEGIN_CS(4);
 145         OUT_CS_REG(R300_VAP_CNTL_STATUS, R300_VAP_TCL_BYPASS);
 146         OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(5) |
 147                 R300_PVS_NUM_CNTLRS(5) |
 148                 R300_PVS_NUM_FPUS(caps->num_vert_fpus) |
 149                 R300_PVS_VF_MAX_VTX_NUM(12));
 150         END_CS;
 151     }
 152
 153     /* Fragment shader setup */
 154     if (caps->is_r500) {
 155         r500_emit_fragment_shader(r300, &r5xx_passthrough_fragment_shader);
 156         r300_emit_rs_block_state(r300, &r5xx_rs_block_clear_state);
 157     } else {
 158         r300_emit_fragment_shader(r300, &r3xx_passthrough_fragment_shader);
 159         r300_emit_rs_block_state(r300, &r3xx_rs_block_clear_state);
 160     }
 161
 162     BEGIN_CS(26);
 163
 164     /* VAP stream control, mapping from input memory to PVS/RS memory */
 165     if (caps->has_tcl) {
 166         OUT_CS_REG(R300_VAP_PROG_STREAM_CNTL_0,
 167             (R300_DATA_TYPE_FLOAT_4 << R300_DATA_TYPE_0_SHIFT) |
 168             ((R300_LAST_VEC | (1 << R300_DST_VEC_LOC_SHIFT) |
 169                 R300_DATA_TYPE_FLOAT_4) << R300_DATA_TYPE_1_SHIFT));
 170     } else {
 171         OUT_CS_REG(R300_VAP_PROG_STREAM_CNTL_0,
 172             (R300_DATA_TYPE_FLOAT_4 << R300_DATA_TYPE_0_SHIFT) |
 173             ((R300_LAST_VEC | (2 << R300_DST_VEC_LOC_SHIFT) |
 174                 R300_DATA_TYPE_FLOAT_4) << R300_DATA_TYPE_1_SHIFT));
 175     }
 176     OUT_CS_REG(R300_VAP_PROG_STREAM_CNTL_EXT_0,
 177             (R300_VAP_SWIZZLE_XYZW << R300_SWIZZLE0_SHIFT) |
 178             (R300_VAP_SWIZZLE_XYZW << R300_SWIZZLE1_SHIFT));
 179
 180     /* VAP format controls */
 181     OUT_CS_REG(R300_VAP_OUTPUT_VTX_FMT_0,
 182             R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT |
 183             R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT);
 184     OUT_CS_REG(R300_VAP_OUTPUT_VTX_FMT_1, 0x0);
 185
 186     /* Disable textures */
 187     OUT_CS_REG(R300_TX_ENABLE, 0x0);
 188
 189     /* The size of the point we're about to draw, in sixths of pixels */
 190     OUT_CS_REG(R300_GA_POINT_SIZE,
 191         ((h * 6)  & R300_POINTSIZE_Y_MASK) |
 192         ((w * 6) << R300_POINTSIZE_X_SHIFT));
 193
 194     /* Vertex size. */
 195     OUT_CS_REG(R300_VAP_VTX_SIZE, 0x8);
 196
 197     /* Packet3 with our point vertex */
 198     OUT_CS_PKT3(R200_3D_DRAW_IMMD_2, 8);
 199     OUT_CS(R300_PRIM_TYPE_POINT | R300_PRIM_WALK_RING |
 200             (1 << R300_PRIM_NUM_VERTICES_SHIFT));
 201     /* Position */
 202     OUT_CS_32F(0.5);
 203     OUT_CS_32F(0.5);
 204     OUT_CS_32F(1.0);
 205     OUT_CS_32F(1.0);
 206     /* Color */
 207     OUT_CS_32F(r);
 208     OUT_CS_32F(g);
 209     OUT_CS_32F(b);
 210     OUT_CS_32F(a);
 211
 212     OUT_CS_REG(R300_RB3D_DSTCACHE_CTLSTAT, 0xA);
 213
 214     END_CS;
 215
 216     r300->dirty_hw++;
 217 }
 218
 219 static void r300_surface_copy(struct pipe_context* pipe,
 220                               struct pipe_surface* dest,
 221                               unsigned destx, unsigned desty,
 222                               struct pipe_surface* src,
 223                               unsigned srcx, unsigned srcy,
 224                               unsigned w, unsigned h)
 225 {
 226     struct r300_context* r300 = r300_context(pipe);
 227     struct r300_capabilities* caps = r300_screen(pipe->screen)->caps;
 228     struct r300_texture* srctex = (struct r300_texture*)src->texture;
 229     struct r300_texture* desttex = (struct r300_texture*)dest->texture;
 230     unsigned pixpitch = srctex->stride / srctex->tex.block.size;
 231     boolean invalid = FALSE;
 232     float fsrcx = srcx, fsrcy = srcy, fdestx = destx, fdesty = desty;
 233     CS_LOCALS(r300);
 234
 235     debug_printf("r300: Copying surface %p at (%d,%d) to %p at (%d, %d),"
 236         " dimensions %dx%d (pixel pitch %d)\n",
 237         src, srcx, srcy, dest, destx, desty, w, h, pixpitch);
 238
 239     if ((srctex->buffer == desttex->buffer) &&
 240             ((destx < srcx + w) || (srcx < destx + w)) &&
 241             ((desty < srcy + h) || (srcy < desty + h))) {
 242 fallback:
 243         debug_printf("r300: Falling back on surface_copy\n");
 244         util_surface_copy(pipe, FALSE, dest, destx, desty, src,
 245                 srcx, srcy, w, h);
 246     }
 247
 248     /* Add our target BOs to the list. */
 249 validate:
 250     if (!r300->winsys->add_buffer(r300->winsys, srctex->buffer,
 251                 RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0)) {
 252         r300->context.flush(&r300->context, 0, NULL);
 253         goto validate;
 254     }
 255     if (!r300->winsys->add_buffer(r300->winsys, desttex->buffer,
 256                 0, RADEON_GEM_DOMAIN_VRAM)) {
 257         r300->context.flush(&r300->context, 0, NULL);
 258         goto validate;
 259     }
 260     if (!r300->winsys->validate(r300->winsys)) {
 261         r300->context.flush(&r300->context, 0, NULL);
 262         if (invalid) {
 263             debug_printf("r300: Stuck in validation loop, gonna fallback.");
 264             goto fallback;
 265         }
 266         invalid = TRUE;
 267         goto validate;
 268     }
 269
 270     r300_surface_setup(r300, desttex, destx, desty, w, h);
 271
 272     /* Setup the texture. */
 273     r300_emit_texture(r300, &r300_sampler_copy_state, srctex, 0);
 274
 275     /* Flush and enable. */
 276     r300_flush_textures(r300);
 277
 278     /* Vertex shader setup */
 279     if (caps->has_tcl) {
 280         r300_emit_vertex_program_code(r300, &r300_passthrough_vertex_shader, 0);
 281     } else {
 282         BEGIN_CS(4);
 283         OUT_CS_REG(R300_VAP_CNTL_STATUS, R300_VAP_TCL_BYPASS);
 284         OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(5) |
 285                 R300_PVS_NUM_CNTLRS(5) |
 286                 R300_PVS_NUM_FPUS(caps->num_vert_fpus) |
 287                 R300_PVS_VF_MAX_VTX_NUM(12));
 288         END_CS;
 289     }
 290
 291     /* Fragment shader setup */
 292     if (caps->is_r500) {
 293         r500_emit_fragment_shader(r300, &r5xx_texture_fragment_shader);
 294         r300_emit_rs_block_state(r300, &r5xx_rs_block_copy_state);
 295     } else {
 296         r300_emit_fragment_shader(r300, &r3xx_texture_fragment_shader);
 297         r300_emit_rs_block_state(r300, &r3xx_rs_block_copy_state);
 298     }
 299
 300     BEGIN_CS(30);
 301     /* VAP stream control, mapping from input memory to PVS/RS memory */
 302     if (caps->has_tcl) {
 303         OUT_CS_REG(R300_VAP_PROG_STREAM_CNTL_0,
 304             (R300_DATA_TYPE_FLOAT_2 << R300_DATA_TYPE_0_SHIFT) |
 305             ((R300_LAST_VEC | (1 << R300_DST_VEC_LOC_SHIFT) |
 306                 R300_DATA_TYPE_FLOAT_2) << R300_DATA_TYPE_1_SHIFT));
 307     } else {
 308         OUT_CS_REG(R300_VAP_PROG_STREAM_CNTL_0,
 309             (R300_DATA_TYPE_FLOAT_2 << R300_DATA_TYPE_0_SHIFT) |
 310             ((R300_LAST_VEC | (6 << R300_DST_VEC_LOC_SHIFT) |
 311                 R300_DATA_TYPE_FLOAT_2) << R300_DATA_TYPE_1_SHIFT));
 312     }
 313     OUT_CS_REG(R300_VAP_PROG_STREAM_CNTL_EXT_0,
 314             (R300_VAP_SWIZZLE_XYZW << R300_SWIZZLE0_SHIFT) |
 315             (R300_VAP_SWIZZLE_XYZW << R300_SWIZZLE1_SHIFT));
 316
 317     /* VAP format controls */
 318     OUT_CS_REG(R300_VAP_OUTPUT_VTX_FMT_0,
 319             R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT);
 320     /* Two components of texture 0 */
 321     OUT_CS_REG(R300_VAP_OUTPUT_VTX_FMT_1, 0x2);
 322
 323     /* Vertex size. */
 324     OUT_CS_REG(R300_VAP_VTX_SIZE, 0x4);
 325
 326     /* Packet3 with our texcoords */
 327     OUT_CS_PKT3(R200_3D_DRAW_IMMD_2, 16);
 328     OUT_CS(R300_PRIM_TYPE_QUADS | R300_PRIM_WALK_RING |
 329             (4 << R300_PRIM_NUM_VERTICES_SHIFT));
 330     /* (x    , y    ) */
 331     OUT_CS_32F(fdestx / dest->width);
 332     OUT_CS_32F(fdesty / dest->height);
 333     OUT_CS_32F(fsrcx  / src->width);
 334     OUT_CS_32F(fsrcy  / src->height);
 335     /* (x    , y + h) */
 336     OUT_CS_32F(fdestx / dest->width);
 337     OUT_CS_32F((fdesty + h) / dest->height);
 338     OUT_CS_32F(fsrcx  / src->width);
 339     OUT_CS_32F((fsrcy  + h) / src->height);
 340     /* (x + w, y + h) */
 341     OUT_CS_32F((fdestx + w) / dest->width);
 342     OUT_CS_32F((fdesty + h) / dest->height);
 343     OUT_CS_32F((fsrcx  + w) / src->width);
 344     OUT_CS_32F((fsrcy  + h) / src->height);
 345     /* (x + w, y    ) */
 346     OUT_CS_32F((fdestx + w) / dest->width);
 347     OUT_CS_32F(fdesty / dest->height);
 348     OUT_CS_32F((fsrcx  + w) / src->width);
 349     OUT_CS_32F(fsrcy  / src->height);
 350
 351     OUT_CS_REG(R300_RB3D_DSTCACHE_CTLSTAT, 0xA);
 352
 353     END_CS;
 354
 355     r300->dirty_hw++;
 356 }
 357
 358 void r300_init_surface_functions(struct r300_context* r300)
 359 {
 360     r300->context.surface_fill = r300_surface_fill;
 361     r300->context.surface_copy = r300_surface_copy;
 362 }