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(26);
  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 - 1) << R300_SCISSORS_X_SHIFT) | ((h - 1) << 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 - 1) + 1440) << R300_SCISSORS_X_SHIFT) |
  67                 (((h - 1) + 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_SEQ(R300_RB3D_COLORPITCH0, 1);
  82     OUT_CS_RELOC(dest->buffer, pixpitch |
  83                  r300_translate_colorformat(dest->tex.format), 0,
  84                  RADEON_GEM_DOMAIN_VRAM, 0);
  85     OUT_CS_REG(RB3D_COLOR_CHANNEL_MASK, 0xf);
  86
  87     END_CS;
  88 }
  89
  90 /* Provides pipe_context's "surface_fill". Commonly used for clearing
  91  * buffers. */
  92 static void r300_surface_fill(struct pipe_context* pipe,
  93                               struct pipe_surface* dest,
  94                               unsigned x, unsigned y,
  95                               unsigned w, unsigned h,
  96                               unsigned color)
  97 {
  98     int i;
  99     float r, g, b, a, depth;
 100     struct r300_context* r300 = r300_context(pipe);
 101     struct r300_capabilities* caps = r300_screen(pipe->screen)->caps;
 102     struct r300_texture* tex = (struct r300_texture*)dest->texture;
 103     unsigned pixpitch = tex->stride / tex->tex.block.size;
 104     boolean invalid = FALSE;
 105     CS_LOCALS(r300);
 106
 107     a = (float)((color >> 24) & 0xff) / 255.0f;
 108     r = (float)((color >> 16) & 0xff) / 255.0f;
 109     g = (float)((color >>  8) & 0xff) / 255.0f;
 110     b = (float)((color >>  0) & 0xff) / 255.0f;
 111     debug_printf("r300: Filling surface %p at (%d,%d),"
 112         " dimensions %dx%d (pixel pitch %d), color 0x%x\n",
 113         dest, x, y, w, h, pixpitch, color);
 114
 115     /* Fallback? */
 116     if (FALSE) {
 117 fallback:
 118         debug_printf("r300: Falling back on surface clear...");
 119         util_surface_fill(pipe, dest, x, y, w, h, color);
 120         return;
 121     }
 122
 123     /* Make sure our target BO is okay. */
 124 validate:
 125     if (!r300->winsys->add_buffer(r300->winsys, tex->buffer,
 126                 0, RADEON_GEM_DOMAIN_VRAM)) {
 127         r300->context.flush(&r300->context, 0, NULL);
 128         goto validate;
 129     }
 130     if (!r300->winsys->validate(r300->winsys)) {
 131         r300->context.flush(&r300->context, 0, NULL);
 132         if (invalid) {
 133             debug_printf("r300: Stuck in validation loop, gonna fallback.");
 134             goto fallback;
 135         }
 136         invalid = TRUE;
 137         goto validate;
 138     }
 139
 140     r300_surface_setup(r300, tex, x, y, w, h);
 141
 142     /* Vertex shader setup */
 143     if (caps->has_tcl) {
 144         r300_emit_vertex_program_code(r300, &r300_passthrough_vertex_shader, 0);
 145     } else {
 146         BEGIN_CS(4);
 147         OUT_CS_REG(R300_VAP_CNTL_STATUS,
 148 #ifdef PIPE_ARCH_BIG_ENDIAN
 149                    R300_VC_32BIT_SWAP |
 150 #endif
 151                    R300_VAP_TCL_BYPASS);
 152         OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(5) |
 153                 R300_PVS_NUM_CNTLRS(5) |
 154                 R300_PVS_NUM_FPUS(caps->num_vert_fpus) |
 155                 R300_PVS_VF_MAX_VTX_NUM(12));
 156         END_CS;
 157     }
 158
 159     /* Fragment shader setup */
 160     if (caps->is_r500) {
 161         r500_emit_fragment_program_code(r300, &r5xx_passthrough_fragment_shader, 0);
 162         r300_emit_rs_block_state(r300, &r5xx_rs_block_clear_state);
 163     } else {
 164         r300_emit_fragment_program_code(r300, &r3xx_passthrough_fragment_shader, 0);
 165         r300_emit_rs_block_state(r300, &r3xx_rs_block_clear_state);
 166     }
 167
 168     BEGIN_CS(26);
 169
 170     /* VAP stream control, mapping from input memory to PVS/RS memory */
 171     if (caps->has_tcl) {
 172         OUT_CS_REG(R300_VAP_PROG_STREAM_CNTL_0,
 173             (R300_DATA_TYPE_FLOAT_4 << R300_DATA_TYPE_0_SHIFT) |
 174             ((R300_LAST_VEC | (1 << R300_DST_VEC_LOC_SHIFT) |
 175                 R300_DATA_TYPE_FLOAT_4) << R300_DATA_TYPE_1_SHIFT));
 176     } else {
 177         OUT_CS_REG(R300_VAP_PROG_STREAM_CNTL_0,
 178             (R300_DATA_TYPE_FLOAT_4 << R300_DATA_TYPE_0_SHIFT) |
 179             ((R300_LAST_VEC | (2 << R300_DST_VEC_LOC_SHIFT) |
 180                 R300_DATA_TYPE_FLOAT_4) << R300_DATA_TYPE_1_SHIFT));
 181     }
 182     OUT_CS_REG(R300_VAP_PROG_STREAM_CNTL_EXT_0,
 183             (R300_VAP_SWIZZLE_XYZW << R300_SWIZZLE0_SHIFT) |
 184             (R300_VAP_SWIZZLE_XYZW << R300_SWIZZLE1_SHIFT));
 185
 186     /* VAP format controls */
 187     OUT_CS_REG(R300_VAP_OUTPUT_VTX_FMT_0,
 188             R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT |
 189             R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT);
 190     OUT_CS_REG(R300_VAP_OUTPUT_VTX_FMT_1, 0x0);
 191
 192     /* Disable textures */
 193     OUT_CS_REG(R300_TX_ENABLE, 0x0);
 194
 195     /* The size of the point we're about to draw, in sixths of pixels */
 196     OUT_CS_REG(R300_GA_POINT_SIZE,
 197         ((h * 6)  & R300_POINTSIZE_Y_MASK) |
 198         ((w * 6) << R300_POINTSIZE_X_SHIFT));
 199
 200     /* Vertex size. */
 201     OUT_CS_REG(R300_VAP_VTX_SIZE, 0x8);
 202
 203     /* Packet3 with our point vertex */
 204     OUT_CS_PKT3(R200_3D_DRAW_IMMD_2, 8);
 205     OUT_CS(R300_PRIM_TYPE_POINT | R300_PRIM_WALK_RING |
 206             (1 << R300_PRIM_NUM_VERTICES_SHIFT));
 207     /* Position */
 208     OUT_CS_32F(0.5);
 209     OUT_CS_32F(0.5);
 210     OUT_CS_32F(1.0);
 211     OUT_CS_32F(1.0);
 212     /* Color */
 213     OUT_CS_32F(r);
 214     OUT_CS_32F(g);
 215     OUT_CS_32F(b);
 216     OUT_CS_32F(a);
 217
 218     OUT_CS_REG(R300_RB3D_DSTCACHE_CTLSTAT, 0xA);
 219
 220     END_CS;
 221
 222     r300->dirty_hw++;
 223 }
 224
 225 static void r300_surface_copy(struct pipe_context* pipe,
 226                               struct pipe_surface* dest,
 227                               unsigned destx, unsigned desty,
 228                               struct pipe_surface* src,
 229                               unsigned srcx, unsigned srcy,
 230                               unsigned w, unsigned h)
 231 {
 232     struct r300_context* r300 = r300_context(pipe);
 233     struct r300_capabilities* caps = r300_screen(pipe->screen)->caps;
 234     struct r300_texture* srctex = (struct r300_texture*)src->texture;
 235     struct r300_texture* desttex = (struct r300_texture*)dest->texture;
 236     unsigned pixpitch = srctex->stride / srctex->tex.block.size;
 237     boolean invalid = FALSE;
 238     float fsrcx = srcx, fsrcy = srcy, fdestx = destx, fdesty = desty;
 239     CS_LOCALS(r300);
 240
 241     debug_printf("r300: Copying surface %p at (%d,%d) to %p at (%d, %d),"
 242         " dimensions %dx%d (pixel pitch %d)\n",
 243         src, srcx, srcy, dest, destx, desty, w, h, pixpitch);
 244
 245     if ((srctex->buffer == desttex->buffer) &&
 246             ((destx < srcx + w) || (srcx < destx + w)) &&
 247             ((desty < srcy + h) || (srcy < desty + h))) {
 248 fallback:
 249         debug_printf("r300: Falling back on surface_copy\n");
 250         util_surface_copy(pipe, FALSE, dest, destx, desty, src,
 251                 srcx, srcy, w, h);
 252     }
 253
 254     /* Add our target BOs to the list. */
 255 validate:
 256     if (!r300->winsys->add_buffer(r300->winsys, srctex->buffer,
 257                 RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0)) {
 258         r300->context.flush(&r300->context, 0, NULL);
 259         goto validate;
 260     }
 261     if (!r300->winsys->add_buffer(r300->winsys, desttex->buffer,
 262                 0, RADEON_GEM_DOMAIN_VRAM)) {
 263         r300->context.flush(&r300->context, 0, NULL);
 264         goto validate;
 265     }
 266     if (!r300->winsys->validate(r300->winsys)) {
 267         r300->context.flush(&r300->context, 0, NULL);
 268         if (invalid) {
 269             debug_printf("r300: Stuck in validation loop, gonna fallback.");
 270             goto fallback;
 271         }
 272         invalid = TRUE;
 273         goto validate;
 274     }
 275
 276     r300_surface_setup(r300, desttex, destx, desty, w, h);
 277
 278     /* Setup the texture. */
 279     r300_emit_texture(r300, &r300_sampler_copy_state, srctex, 0);
 280
 281     /* Flush and enable. */
 282     r300_flush_textures(r300);
 283
 284     /* Vertex shader setup */
 285     if (caps->has_tcl) {
 286         r300_emit_vertex_program_code(r300, &r300_passthrough_vertex_shader, 0);
 287     } else {
 288         BEGIN_CS(4);
 289         OUT_CS_REG(R300_VAP_CNTL_STATUS,
 290 #ifdef PIPE_ARCH_BIG_ENDIAN
 291                    R300_VC_32BIT_SWAP |
 292 #endif
 293                    R300_VAP_TCL_BYPASS);
 294         OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(5) |
 295                 R300_PVS_NUM_CNTLRS(5) |
 296                 R300_PVS_NUM_FPUS(caps->num_vert_fpus) |
 297                 R300_PVS_VF_MAX_VTX_NUM(12));
 298         END_CS;
 299     }
 300
 301     /* Fragment shader setup */
 302     if (caps->is_r500) {
 303         r500_emit_fragment_program_code(r300, &r5xx_texture_fragment_shader, 0);
 304         r300_emit_rs_block_state(r300, &r5xx_rs_block_copy_state);
 305     } else {
 306         r300_emit_fragment_program_code(r300, &r3xx_texture_fragment_shader, 0);
 307         r300_emit_rs_block_state(r300, &r3xx_rs_block_copy_state);
 308     }
 309
 310     BEGIN_CS(30);
 311     /* VAP stream control, mapping from input memory to PVS/RS memory */
 312     if (caps->has_tcl) {
 313         OUT_CS_REG(R300_VAP_PROG_STREAM_CNTL_0,
 314             (R300_DATA_TYPE_FLOAT_2 << R300_DATA_TYPE_0_SHIFT) |
 315             ((R300_LAST_VEC | (1 << R300_DST_VEC_LOC_SHIFT) |
 316                 R300_DATA_TYPE_FLOAT_2) << R300_DATA_TYPE_1_SHIFT));
 317     } else {
 318         OUT_CS_REG(R300_VAP_PROG_STREAM_CNTL_0,
 319             (R300_DATA_TYPE_FLOAT_2 << R300_DATA_TYPE_0_SHIFT) |
 320             ((R300_LAST_VEC | (6 << R300_DST_VEC_LOC_SHIFT) |
 321                 R300_DATA_TYPE_FLOAT_2) << R300_DATA_TYPE_1_SHIFT));
 322     }
 323     OUT_CS_REG(R300_VAP_PROG_STREAM_CNTL_EXT_0,
 324             (R300_VAP_SWIZZLE_XYZW << R300_SWIZZLE0_SHIFT) |
 325             (R300_VAP_SWIZZLE_XYZW << R300_SWIZZLE1_SHIFT));
 326
 327     /* VAP format controls */
 328     OUT_CS_REG(R300_VAP_OUTPUT_VTX_FMT_0,
 329             R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT);
 330     /* Two components of texture 0 */
 331     OUT_CS_REG(R300_VAP_OUTPUT_VTX_FMT_1, 0x2);
 332
 333     /* Vertex size. */
 334     OUT_CS_REG(R300_VAP_VTX_SIZE, 0x4);
 335
 336     /* Packet3 with our texcoords */
 337     OUT_CS_PKT3(R200_3D_DRAW_IMMD_2, 16);
 338     OUT_CS(R300_PRIM_TYPE_QUADS | R300_PRIM_WALK_RING |
 339             (4 << R300_PRIM_NUM_VERTICES_SHIFT));
 340     /* (x    , y    ) */
 341     OUT_CS_32F(fdestx / dest->width);
 342     OUT_CS_32F(fdesty / dest->height);
 343     OUT_CS_32F(fsrcx  / src->width);
 344     OUT_CS_32F(fsrcy  / src->height);
 345     /* (x    , y + h) */
 346     OUT_CS_32F(fdestx / dest->width);
 347     OUT_CS_32F((fdesty + h) / dest->height);
 348     OUT_CS_32F(fsrcx  / src->width);
 349     OUT_CS_32F((fsrcy  + h) / src->height);
 350     /* (x + w, y + h) */
 351     OUT_CS_32F((fdestx + w) / dest->width);
 352     OUT_CS_32F((fdesty + h) / dest->height);
 353     OUT_CS_32F((fsrcx  + w) / src->width);
 354     OUT_CS_32F((fsrcy  + h) / src->height);
 355     /* (x + w, y    ) */
 356     OUT_CS_32F((fdestx + w) / dest->width);
 357     OUT_CS_32F(fdesty / dest->height);
 358     OUT_CS_32F((fsrcx  + w) / src->width);
 359     OUT_CS_32F(fsrcy  / src->height);
 360
 361     OUT_CS_REG(R300_RB3D_DSTCACHE_CTLSTAT, 0xA);
 362
 363     END_CS;
 364
 365     r300->dirty_hw++;
 366 }
 367
 368 void r300_init_surface_functions(struct r300_context* r300)
 369 {
 370     r300->context.surface_fill = r300_surface_fill;
 371     r300->context.surface_copy = r300_surface_copy;
 372 }