brw_blorp_blit.cpp revision 19e9b24626c2b9d7abef054d57bb2a52106c545b
1/* 2 * Copyright © 2012 Intel Corporation 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice (including the next 12 * paragraph) shall be included in all copies or substantial portions of the 13 * Software. 14 * 15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS 21 * IN THE SOFTWARE. 22 */ 23 24#include "main/teximage.h" 25 26#include "glsl/ralloc.h" 27 28#include "intel_fbo.h" 29 30#include "brw_blorp.h" 31#include "brw_context.h" 32#include "brw_eu.h" 33#include "brw_state.h" 34 35 36/** 37 * Helper function for handling mirror image blits. 38 * 39 * If coord0 > coord1, swap them and invert the "mirror" boolean. 40 */ 41static inline void 42fixup_mirroring(bool &mirror, GLint &coord0, GLint &coord1) 43{ 44 if (coord0 > coord1) { 45 mirror = !mirror; 46 GLint tmp = coord0; 47 coord0 = coord1; 48 coord1 = tmp; 49 } 50} 51 52 53static bool 54try_blorp_blit(struct intel_context *intel, 55 GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1, 56 GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1, 57 GLenum filter, GLbitfield buffer_bit) 58{ 59 struct gl_context *ctx = &intel->ctx; 60 61 /* Sync up the state of window system buffers. We need to do this before 62 * we go looking for the buffers. 63 */ 64 intel_prepare_render(intel); 65 66 /* Find buffers */ 67 const struct gl_framebuffer *read_fb = ctx->ReadBuffer; 68 const struct gl_framebuffer *draw_fb = ctx->DrawBuffer; 69 struct gl_renderbuffer *src_rb; 70 struct gl_renderbuffer *dst_rb; 71 switch (buffer_bit) { 72 case GL_COLOR_BUFFER_BIT: 73 src_rb = read_fb->_ColorReadBuffer; 74 dst_rb = 75 draw_fb->Attachment[ 76 draw_fb->_ColorDrawBufferIndexes[0]].Renderbuffer; 77 break; 78 case GL_DEPTH_BUFFER_BIT: 79 src_rb = read_fb->Attachment[BUFFER_DEPTH].Renderbuffer; 80 dst_rb = draw_fb->Attachment[BUFFER_DEPTH].Renderbuffer; 81 break; 82 case GL_STENCIL_BUFFER_BIT: 83 src_rb = read_fb->Attachment[BUFFER_STENCIL].Renderbuffer; 84 dst_rb = draw_fb->Attachment[BUFFER_STENCIL].Renderbuffer; 85 break; 86 default: 87 assert(false); 88 } 89 90 /* Validate source */ 91 if (!src_rb) return false; 92 struct intel_renderbuffer *src_irb = intel_renderbuffer(src_rb); 93 struct intel_mipmap_tree *src_mt = src_irb->mt; 94 if (!src_mt) return false; 95 if (buffer_bit == GL_STENCIL_BUFFER_BIT && src_mt->stencil_mt) 96 src_mt = src_mt->stencil_mt; 97 switch (src_mt->format) { 98 case MESA_FORMAT_ARGB8888: 99 case MESA_FORMAT_X8_Z24: 100 case MESA_FORMAT_S8: 101 break; /* Supported */ 102 default: 103 /* Unsupported format. 104 * 105 * TODO: need to support all formats that are allowed as multisample 106 * render targets. 107 */ 108 return false; 109 } 110 111 /* Validate destination */ 112 if (!dst_rb) return false; 113 struct intel_renderbuffer *dst_irb = intel_renderbuffer(dst_rb); 114 struct intel_mipmap_tree *dst_mt = dst_irb->mt; 115 if (!dst_mt) return false; 116 if (buffer_bit == GL_STENCIL_BUFFER_BIT && dst_mt->stencil_mt) 117 dst_mt = dst_mt->stencil_mt; 118 switch (dst_mt->format) { 119 case MESA_FORMAT_ARGB8888: 120 case MESA_FORMAT_X8_Z24: 121 case MESA_FORMAT_S8: 122 break; /* Supported */ 123 default: 124 /* Unsupported format. 125 * 126 * TODO: need to support all formats that are allowed as multisample 127 * render targets. 128 */ 129 return false; 130 } 131 132 /* Account for the fact that in the system framebuffer, the origin is at 133 * the lower left. 134 */ 135 if (read_fb->Name == 0) { 136 srcY0 = read_fb->Height - srcY0; 137 srcY1 = read_fb->Height - srcY1; 138 } 139 if (draw_fb->Name == 0) { 140 dstY0 = draw_fb->Height - dstY0; 141 dstY1 = draw_fb->Height - dstY1; 142 } 143 144 /* Detect if the blit needs to be mirrored */ 145 bool mirror_x = false, mirror_y = false; 146 fixup_mirroring(mirror_x, srcX0, srcX1); 147 fixup_mirroring(mirror_x, dstX0, dstX1); 148 fixup_mirroring(mirror_y, srcY0, srcY1); 149 fixup_mirroring(mirror_y, dstY0, dstY1); 150 151 /* Make sure width and height match */ 152 GLsizei width = srcX1 - srcX0; 153 GLsizei height = srcY1 - srcY0; 154 if (width != dstX1 - dstX0) return false; 155 if (height != dstY1 - dstY0) return false; 156 157 /* Make sure width and height don't need to be clipped or scissored. 158 * TODO: support clipping and scissoring. 159 */ 160 if (srcX0 < 0 || (GLuint) srcX1 > read_fb->Width) return false; 161 if (srcY0 < 0 || (GLuint) srcY1 > read_fb->Height) return false; 162 if (dstX0 < 0 || (GLuint) dstX1 > draw_fb->Width) return false; 163 if (dstY0 < 0 || (GLuint) dstY1 > draw_fb->Height) return false; 164 if (ctx->Scissor.Enabled) return false; 165 166 /* Get ready to blit. This includes depth resolving the src and dst 167 * buffers if necessary. 168 */ 169 intel_renderbuffer_resolve_depth(intel, src_irb); 170 intel_renderbuffer_resolve_depth(intel, dst_irb); 171 172 /* Do the blit */ 173 brw_blorp_blit_params params(src_mt, dst_mt, 174 srcX0, srcY0, dstX0, dstY0, dstX1, dstY1, 175 mirror_x, mirror_y); 176 params.exec(intel); 177 178 /* Mark the dst buffer as needing a HiZ resolve if necessary. */ 179 intel_renderbuffer_set_needs_hiz_resolve(dst_irb); 180 181 return true; 182} 183 184GLbitfield 185brw_blorp_framebuffer(struct intel_context *intel, 186 GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1, 187 GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1, 188 GLbitfield mask, GLenum filter) 189{ 190 /* BLORP is only supported on Gen6. TODO: implement on Gen7. */ 191 if (intel->gen != 6) 192 return mask; 193 194 static GLbitfield buffer_bits[] = { 195 GL_COLOR_BUFFER_BIT, 196 GL_DEPTH_BUFFER_BIT, 197 GL_STENCIL_BUFFER_BIT, 198 }; 199 200 for (unsigned int i = 0; i < ARRAY_SIZE(buffer_bits); ++i) { 201 if ((mask & buffer_bits[i]) && 202 try_blorp_blit(intel, 203 srcX0, srcY0, srcX1, srcY1, 204 dstX0, dstY0, dstX1, dstY1, 205 filter, buffer_bits[i])) { 206 mask &= ~buffer_bits[i]; 207 } 208 } 209 210 return mask; 211} 212 213/** 214 * Generator for WM programs used in BLORP blits. 215 * 216 * The bulk of the work done by the WM program is to wrap and unwrap the 217 * coordinate transformations used by the hardware to store surfaces in 218 * memory. The hardware transforms a pixel location (X, Y, S) (where S is the 219 * sample index for a multisampled surface) to a memory offset by the 220 * following formulas: 221 * 222 * offset = tile(tiling_format, encode_msaa(num_samples, X, Y, S)) 223 * (X, Y, S) = decode_msaa(num_samples, detile(tiling_format, offset)) 224 * 225 * For a single-sampled surface, encode_msaa() and decode_msaa are the 226 * identity function: 227 * 228 * encode_msaa(1, X, Y, 0) = (X, Y) 229 * decode_msaa(1, X, Y) = (X, Y, 0) 230 * 231 * For a 4x multisampled surface, encode_msaa() embeds the sample number into 232 * bit 1 of the X and Y coordinates: 233 * 234 * encode_msaa(4, X, Y, S) = (X', Y') 235 * where X' = (X & ~0b1) << 1 | (S & 0b1) << 1 | (X & 0b1) 236 * Y' = (Y & ~0b1 ) << 1 | (S & 0b10) | (Y & 0b1) 237 * decode_msaa(4, X, Y) = (X', Y', S) 238 * where X' = (X & ~0b11) >> 1 | (X & 0b1) 239 * Y' = (Y & ~0b11) >> 1 | (Y & 0b1) 240 * S = (Y & 0b10) | (X & 0b10) >> 1 241 * 242 * For X tiling, tile() combines together the low-order bits of the X and Y 243 * coordinates in the pattern 0byyyxxxxxxxxx, creating 4k tiles that are 512 244 * bytes wide and 8 rows high: 245 * 246 * tile(x_tiled, X, Y) = A 247 * where A = tile_num << 12 | offset 248 * tile_num = (Y >> 3) * tile_pitch + (X' >> 9) 249 * offset = (Y & 0b111) << 9 250 * | (X & 0b111111111) 251 * X' = X * cpp 252 * detile(x_tiled, A) = (X, Y) 253 * where X = X' / cpp 254 * Y = (tile_num / tile_pitch) << 3 255 * | (A & 0b111000000000) >> 9 256 * X' = (tile_num % tile_pitch) << 9 257 * | (A & 0b111111111) 258 * 259 * (In all tiling formulas, cpp is the number of bytes occupied by a single 260 * sample ("chars per pixel"), and tile_pitch is the number of 4k tiles 261 * required to fill the width of the surface). 262 * 263 * For Y tiling, tile() combines together the low-order bits of the X and Y 264 * coordinates in the pattern 0bxxxyyyyyxxxx, creating 4k tiles that are 128 265 * bytes wide and 32 rows high: 266 * 267 * tile(y_tiled, X, Y) = A 268 * where A = tile_num << 12 | offset 269 * tile_num = (Y >> 5) * tile_pitch + (X' >> 7) 270 * offset = (X' & 0b1110000) << 5 271 * | (Y' & 0b11111) << 4 272 * | (X' & 0b1111) 273 * X' = X * cpp 274 * detile(y_tiled, A) = (X, Y) 275 * where X = X' / cpp 276 * Y = (tile_num / tile_pitch) << 5 277 * | (A & 0b111110000) >> 4 278 * X' = (tile_num % tile_pitch) << 7 279 * | (A & 0b111000000000) >> 5 280 * | (A & 0b1111) 281 * 282 * For W tiling, tile() combines together the low-order bits of the X and Y 283 * coordinates in the pattern 0bxxxyyyyxyxyx, creating 4k tiles that are 64 284 * bytes wide and 64 rows high (note that W tiling is only used for stencil 285 * buffers, which always have cpp = 1): 286 * 287 * tile(w_tiled, X, Y) = A 288 * where A = tile_num << 12 | offset 289 * tile_num = (Y >> 6) * tile_pitch + (X' >> 6) 290 * offset = (X' & 0b111000) << 6 291 * | (Y & 0b111100) << 3 292 * | (X' & 0b100) << 2 293 * | (Y & 0b10) << 2 294 * | (X' & 0b10) << 1 295 * | (Y & 0b1) << 1 296 * | (X' & 0b1) 297 * X' = X * cpp = X 298 * detile(w_tiled, A) = (X, Y) 299 * where X = X' / cpp = X' 300 * Y = (tile_num / tile_pitch) << 6 301 * | (A & 0b111100000) >> 3 302 * | (A & 0b1000) >> 2 303 * | (A & 0b10) >> 1 304 * X' = (tile_num % tile_pitch) << 6 305 * | (A & 0b111000000000) >> 6 306 * | (A & 0b10000) >> 2 307 * | (A & 0b100) >> 1 308 * | (A & 0b1) 309 * 310 * Finally, for a non-tiled surface, tile() simply combines together the X and 311 * Y coordinates in the natural way: 312 * 313 * tile(untiled, X, Y) = A 314 * where A = Y * pitch + X' 315 * X' = X * cpp 316 * detile(untiled, A) = (X, Y) 317 * where X = X' / cpp 318 * Y = A / pitch 319 * X' = A % pitch 320 * 321 * (In these formulas, pitch is the number of bytes occupied by a single row 322 * of samples). 323 */ 324class brw_blorp_blit_program 325{ 326public: 327 brw_blorp_blit_program(struct brw_context *brw, 328 const brw_blorp_blit_prog_key *key); 329 ~brw_blorp_blit_program(); 330 331 const GLuint *compile(struct brw_context *brw, GLuint *program_size); 332 333 brw_blorp_prog_data prog_data; 334 335private: 336 void alloc_regs(); 337 void alloc_push_const_regs(int base_reg); 338 void compute_frag_coords(); 339 void translate_tiling(bool old_tiled_w, bool new_tiled_w); 340 void encode_msaa(unsigned num_samples); 341 void decode_msaa(unsigned num_samples); 342 void kill_if_outside_dst_rect(); 343 void translate_dst_to_src(); 344 void single_to_blend(); 345 void sample(); 346 void texel_fetch(); 347 void texture_lookup(GLuint msg_type, 348 struct brw_reg mrf_u, struct brw_reg mrf_v); 349 void render_target_write(); 350 351 void *mem_ctx; 352 struct brw_context *brw; 353 const brw_blorp_blit_prog_key *key; 354 struct brw_compile func; 355 356 /* Thread dispatch header */ 357 struct brw_reg R0; 358 359 /* Pixel X/Y coordinates (always in R1). */ 360 struct brw_reg R1; 361 362 /* Push constants */ 363 struct brw_reg dst_x0; 364 struct brw_reg dst_x1; 365 struct brw_reg dst_y0; 366 struct brw_reg dst_y1; 367 struct { 368 struct brw_reg multiplier; 369 struct brw_reg offset; 370 } x_transform, y_transform; 371 372 /* Data returned from texture lookup (4 vec16's) */ 373 struct brw_reg Rdata; 374 375 /* X coordinates. We have two of them so that we can perform coordinate 376 * transformations easily. 377 */ 378 struct brw_reg x_coords[2]; 379 380 /* Y coordinates. We have two of them so that we can perform coordinate 381 * transformations easily. 382 */ 383 struct brw_reg y_coords[2]; 384 385 /* Which element of x_coords and y_coords is currently in use. 386 */ 387 int xy_coord_index; 388 389 /* True if, at the point in the program currently being compiled, the 390 * sample index is known to be zero. 391 */ 392 bool s_is_zero; 393 394 /* Register storing the sample index when s_is_zero is false. */ 395 struct brw_reg sample_index; 396 397 /* Temporaries */ 398 struct brw_reg t1; 399 struct brw_reg t2; 400 401 /* M2-3: u coordinate */ 402 GLuint base_mrf; 403 struct brw_reg mrf_u_float; 404 405 /* M4-5: v coordinate */ 406 struct brw_reg mrf_v_float; 407}; 408 409brw_blorp_blit_program::brw_blorp_blit_program( 410 struct brw_context *brw, 411 const brw_blorp_blit_prog_key *key) 412 : mem_ctx(ralloc_context(NULL)), 413 brw(brw), 414 key(key) 415{ 416 brw_init_compile(brw, &func, mem_ctx); 417} 418 419brw_blorp_blit_program::~brw_blorp_blit_program() 420{ 421 ralloc_free(mem_ctx); 422} 423 424const GLuint * 425brw_blorp_blit_program::compile(struct brw_context *brw, 426 GLuint *program_size) 427{ 428 /* Sanity checks */ 429 if (key->src_tiled_w) { 430 /* If the source image is W tiled, then tex_samples must be 0. 431 * Otherwise, after conversion between W and Y tiling, there's no 432 * guarantee that the sample index will be 0. 433 */ 434 assert(key->tex_samples == 0); 435 } 436 437 if (key->dst_tiled_w) { 438 /* If the destination image is W tiled, then dst_samples must be 0. 439 * Otherwise, after conversion between W and Y tiling, there's no 440 * guarantee that all samples corresponding to a single pixel will still 441 * be together. 442 */ 443 assert(key->rt_samples == 0); 444 } 445 446 if (key->blend) { 447 /* We are blending, which means we'll be using a SAMPLE message, which 448 * causes the hardware to pick up the all of the samples corresponding 449 * to this pixel and average them together. Since we'll be relying on 450 * the hardware to find all of the samples and combine them together, 451 * the surface state for the texture must be configured with the correct 452 * tiling and sample count. 453 */ 454 assert(!key->src_tiled_w); 455 assert(key->tex_samples == key->src_samples); 456 assert(key->tex_samples > 0); 457 } 458 459 brw_set_compression_control(&func, BRW_COMPRESSION_NONE); 460 461 alloc_regs(); 462 compute_frag_coords(); 463 464 /* Render target and texture hardware don't support W tiling. */ 465 const bool rt_tiled_w = false; 466 const bool tex_tiled_w = false; 467 468 /* The address that data will be written to is determined by the 469 * coordinates supplied to the WM thread and the tiling and sample count of 470 * the render target, according to the formula: 471 * 472 * (X, Y, S) = decode_msaa(rt_samples, detile(rt_tiling, offset)) 473 * 474 * If the actual tiling and sample count of the destination surface are not 475 * the same as the configuration of the render target, then these 476 * coordinates are wrong and we have to adjust them to compensate for the 477 * difference. 478 */ 479 if (rt_tiled_w != key->dst_tiled_w || 480 key->rt_samples != key->dst_samples) { 481 encode_msaa(key->rt_samples); 482 /* Now (X, Y) = detile(rt_tiling, offset) */ 483 translate_tiling(rt_tiled_w, key->dst_tiled_w); 484 /* Now (X, Y) = detile(dst_tiling, offset) */ 485 decode_msaa(key->dst_samples); 486 } 487 488 /* Now (X, Y, S) = decode_msaa(dst_samples, detile(dst_tiling, offset)). 489 * 490 * That is: X, Y and S now contain the true coordinates and sample index of 491 * the data that the WM thread should output. 492 * 493 * If we need to kill pixels that are outside the destination rectangle, 494 * now is the time to do it. 495 */ 496 497 if (key->use_kill) 498 kill_if_outside_dst_rect(); 499 500 /* Next, apply a translation to obtain coordinates in the source image. */ 501 translate_dst_to_src(); 502 503 /* If the source image is not multisampled, then we want to fetch sample 504 * number 0, because that's the only sample there is. 505 */ 506 if (key->src_samples == 0) 507 s_is_zero = true; 508 509 /* X, Y, and S are now the coordinates of the pixel in the source image 510 * that we want to texture from. Exception: if we are blending, then S is 511 * irrelevant, because we are going to fetch all samples. 512 */ 513 if (key->blend) { 514 single_to_blend(); 515 sample(); 516 } else { 517 /* We aren't blending, which means we just want to fetch a single sample 518 * from the source surface. The address that we want to fetch from is 519 * related to the X, Y and S values according to the formula: 520 * 521 * (X, Y, S) = decode_msaa(src_samples, detile(src_tiling, offset)). 522 * 523 * If the actual tiling and sample count of the source surface are not 524 * the same as the configuration of the texture, then we need to adjust 525 * the coordinates to compensate for the difference. 526 */ 527 if (tex_tiled_w != key->src_tiled_w || 528 key->tex_samples != key->src_samples) { 529 encode_msaa(key->src_samples); 530 /* Now (X, Y) = detile(src_tiling, offset) */ 531 translate_tiling(key->src_tiled_w, tex_tiled_w); 532 /* Now (X, Y) = detile(tex_tiling, offset) */ 533 decode_msaa(key->tex_samples); 534 } 535 536 /* Now (X, Y, S) = decode_msaa(tex_samples, detile(tex_tiling, offset)). 537 * 538 * In other words: X, Y, and S now contain values which, when passed to 539 * the texturing unit, will cause data to be read from the correct 540 * memory location. So we can fetch the texel now. 541 */ 542 texel_fetch(); 543 } 544 545 /* Finally, write the fetched (or blended) value to the render target and 546 * terminate the thread. 547 */ 548 render_target_write(); 549 return brw_get_program(&func, program_size); 550} 551 552void 553brw_blorp_blit_program::alloc_push_const_regs(int base_reg) 554{ 555#define CONST_LOC(name) offsetof(brw_blorp_wm_push_constants, name) 556#define ALLOC_REG(name) \ 557 this->name = \ 558 brw_uw1_reg(BRW_GENERAL_REGISTER_FILE, base_reg, CONST_LOC(name) / 2) 559 560 ALLOC_REG(dst_x0); 561 ALLOC_REG(dst_x1); 562 ALLOC_REG(dst_y0); 563 ALLOC_REG(dst_y1); 564 ALLOC_REG(x_transform.multiplier); 565 ALLOC_REG(x_transform.offset); 566 ALLOC_REG(y_transform.multiplier); 567 ALLOC_REG(y_transform.offset); 568#undef CONST_LOC 569#undef ALLOC_REG 570} 571 572void 573brw_blorp_blit_program::alloc_regs() 574{ 575 int reg = 0; 576 this->R0 = retype(brw_vec8_grf(reg++, 0), BRW_REGISTER_TYPE_UW); 577 this->R1 = retype(brw_vec8_grf(reg++, 0), BRW_REGISTER_TYPE_UW); 578 prog_data.first_curbe_grf = reg; 579 alloc_push_const_regs(reg); 580 reg += BRW_BLORP_NUM_PUSH_CONST_REGS; 581 this->Rdata = vec16(brw_vec8_grf(reg, 0)); reg += 8; 582 for (int i = 0; i < 2; ++i) { 583 this->x_coords[i] 584 = vec16(retype(brw_vec8_grf(reg++, 0), BRW_REGISTER_TYPE_UW)); 585 this->y_coords[i] 586 = vec16(retype(brw_vec8_grf(reg++, 0), BRW_REGISTER_TYPE_UW)); 587 } 588 this->xy_coord_index = 0; 589 this->sample_index 590 = vec16(retype(brw_vec8_grf(reg++, 0), BRW_REGISTER_TYPE_UW)); 591 this->t1 = vec16(retype(brw_vec8_grf(reg++, 0), BRW_REGISTER_TYPE_UW)); 592 this->t2 = vec16(retype(brw_vec8_grf(reg++, 0), BRW_REGISTER_TYPE_UW)); 593 594 int mrf = 2; 595 this->base_mrf = mrf; 596 this->mrf_u_float = vec16(brw_message_reg(mrf)); mrf += 2; 597 this->mrf_v_float = vec16(brw_message_reg(mrf)); mrf += 2; 598} 599 600/* In the code that follows, X and Y can be used to quickly refer to the 601 * active elements of x_coords and y_coords, and Xp and Yp ("X prime" and "Y 602 * prime") to the inactive elements. 603 * 604 * S can be used to quickly refer to sample_index. 605 */ 606#define X x_coords[xy_coord_index] 607#define Y y_coords[xy_coord_index] 608#define Xp x_coords[!xy_coord_index] 609#define Yp y_coords[!xy_coord_index] 610#define S sample_index 611 612/* Quickly swap the roles of (X, Y) and (Xp, Yp). Saves us from having to do 613 * MOVs to transfor (Xp, Yp) to (X, Y) after a coordinate transformation. 614 */ 615#define SWAP_XY_AND_XPYP() xy_coord_index = !xy_coord_index; 616 617/** 618 * Emit code to compute the X and Y coordinates of the pixels being rendered 619 * by this WM invocation. 620 * 621 * Assuming the render target is set up for Y tiling, these (X, Y) values are 622 * related to the address offset where outputs will be written by the formula: 623 * 624 * (X, Y, S) = decode_msaa(detile(offset)). 625 * 626 * (See brw_blorp_blit_program). 627 */ 628void 629brw_blorp_blit_program::compute_frag_coords() 630{ 631 /* R1.2[15:0] = X coordinate of upper left pixel of subspan 0 (pixel 0) 632 * R1.3[15:0] = X coordinate of upper left pixel of subspan 1 (pixel 4) 633 * R1.4[15:0] = X coordinate of upper left pixel of subspan 2 (pixel 8) 634 * R1.5[15:0] = X coordinate of upper left pixel of subspan 3 (pixel 12) 635 * 636 * Pixels within a subspan are laid out in this arrangement: 637 * 0 1 638 * 2 3 639 * 640 * So, to compute the coordinates of each pixel, we need to read every 2nd 641 * 16-bit value (vstride=2) from R1, starting at the 4th 16-bit value 642 * (suboffset=4), and duplicate each value 4 times (hstride=0, width=4). 643 * In other words, the data we want to access is R1.4<2;4,0>UW. 644 * 645 * Then, we need to add the repeating sequence (0, 1, 0, 1, ...) to the 646 * result, since pixels n+1 and n+3 are in the right half of the subspan. 647 */ 648 brw_ADD(&func, X, stride(suboffset(R1, 4), 2, 4, 0), brw_imm_v(0x10101010)); 649 650 /* Similarly, Y coordinates for subspans come from R1.2[31:16] through 651 * R1.5[31:16], so to get pixel Y coordinates we need to start at the 5th 652 * 16-bit value instead of the 4th (R1.5<2;4,0>UW instead of 653 * R1.4<2;4,0>UW). 654 * 655 * And we need to add the repeating sequence (0, 0, 1, 1, ...), since 656 * pixels n+2 and n+3 are in the bottom half of the subspan. 657 */ 658 brw_ADD(&func, Y, stride(suboffset(R1, 5), 2, 4, 0), brw_imm_v(0x11001100)); 659 660 /* Since we always run the WM in a mode that causes a single fragment 661 * dispatch per pixel, it's not meaningful to compute a sample value. Just 662 * set it to 0. 663 */ 664 s_is_zero = true; 665} 666 667/** 668 * Emit code to compensate for the difference between Y and W tiling. 669 * 670 * This code modifies the X and Y coordinates according to the formula: 671 * 672 * (X', Y') = detile(new_tiling, tile(old_tiling, X, Y)) 673 * 674 * (See brw_blorp_blit_program). 675 * 676 * It can only translate between W and Y tiling, so new_tiling and old_tiling 677 * are booleans where true represents W tiling and false represents Y tiling. 678 */ 679void 680brw_blorp_blit_program::translate_tiling(bool old_tiled_w, bool new_tiled_w) 681{ 682 if (old_tiled_w == new_tiled_w) 683 return; 684 685 if (new_tiled_w) { 686 /* Given X and Y coordinates that describe an address using Y tiling, 687 * translate to the X and Y coordinates that describe the same address 688 * using W tiling. 689 * 690 * If we break down the low order bits of X and Y, using a 691 * single letter to represent each low-order bit: 692 * 693 * X = A << 7 | 0bBCDEFGH 694 * Y = J << 5 | 0bKLMNP (1) 695 * 696 * Then we can apply the Y tiling formula to see the memory offset being 697 * addressed: 698 * 699 * offset = (J * tile_pitch + A) << 12 | 0bBCDKLMNPEFGH (2) 700 * 701 * If we apply the W detiling formula to this memory location, that the 702 * corresponding X' and Y' coordinates are: 703 * 704 * X' = A << 6 | 0bBCDPFH (3) 705 * Y' = J << 6 | 0bKLMNEG 706 * 707 * Combining (1) and (3), we see that to transform (X, Y) to (X', Y'), 708 * we need to make the following computation: 709 * 710 * X' = (X & ~0b1011) >> 1 | (Y & 0b1) << 2 | X & 0b1 (4) 711 * Y' = (Y & ~0b1) << 1 | (X & 0b1000) >> 2 | (X & 0b10) >> 1 712 */ 713 brw_AND(&func, t1, X, brw_imm_uw(0xfff4)); /* X & ~0b1011 */ 714 brw_SHR(&func, t1, t1, brw_imm_uw(1)); /* (X & ~0b1011) >> 1 */ 715 brw_AND(&func, t2, Y, brw_imm_uw(1)); /* Y & 0b1 */ 716 brw_SHL(&func, t2, t2, brw_imm_uw(2)); /* (Y & 0b1) << 2 */ 717 brw_OR(&func, t1, t1, t2); /* (X & ~0b1011) >> 1 | (Y & 0b1) << 2 */ 718 brw_AND(&func, t2, X, brw_imm_uw(1)); /* X & 0b1 */ 719 brw_OR(&func, Xp, t1, t2); 720 brw_AND(&func, t1, Y, brw_imm_uw(0xfffe)); /* Y & ~0b1 */ 721 brw_SHL(&func, t1, t1, brw_imm_uw(1)); /* (Y & ~0b1) << 1 */ 722 brw_AND(&func, t2, X, brw_imm_uw(8)); /* X & 0b1000 */ 723 brw_SHR(&func, t2, t2, brw_imm_uw(2)); /* (X & 0b1000) >> 2 */ 724 brw_OR(&func, t1, t1, t2); /* (Y & ~0b1) << 1 | (X & 0b1000) >> 2 */ 725 brw_AND(&func, t2, X, brw_imm_uw(2)); /* X & 0b10 */ 726 brw_SHR(&func, t2, t2, brw_imm_uw(1)); /* (X & 0b10) >> 1 */ 727 brw_OR(&func, Yp, t1, t2); 728 SWAP_XY_AND_XPYP(); 729 } else { 730 /* Applying the same logic as above, but in reverse, we obtain the 731 * formulas: 732 * 733 * X' = (X & ~0b101) << 1 | (Y & 0b10) << 2 | (Y & 0b1) << 1 | X & 0b1 734 * Y' = (Y & ~0b11) >> 1 | (X & 0b100) >> 2 735 */ 736 brw_AND(&func, t1, X, brw_imm_uw(0xfffa)); /* X & ~0b101 */ 737 brw_SHL(&func, t1, t1, brw_imm_uw(1)); /* (X & ~0b101) << 1 */ 738 brw_AND(&func, t2, Y, brw_imm_uw(2)); /* Y & 0b10 */ 739 brw_SHL(&func, t2, t2, brw_imm_uw(2)); /* (Y & 0b10) << 2 */ 740 brw_OR(&func, t1, t1, t2); /* (X & ~0b101) << 1 | (Y & 0b10) << 2 */ 741 brw_AND(&func, t2, Y, brw_imm_uw(1)); /* Y & 0b1 */ 742 brw_SHL(&func, t2, t2, brw_imm_uw(1)); /* (Y & 0b1) << 1 */ 743 brw_OR(&func, t1, t1, t2); /* (X & ~0b101) << 1 | (Y & 0b10) << 2 744 | (Y & 0b1) << 1 */ 745 brw_AND(&func, t2, X, brw_imm_uw(1)); /* X & 0b1 */ 746 brw_OR(&func, Xp, t1, t2); 747 brw_AND(&func, t1, Y, brw_imm_uw(0xfffc)); /* Y & ~0b11 */ 748 brw_SHR(&func, t1, t1, brw_imm_uw(1)); /* (Y & ~0b11) >> 1 */ 749 brw_AND(&func, t2, X, brw_imm_uw(4)); /* X & 0b100 */ 750 brw_SHR(&func, t2, t2, brw_imm_uw(2)); /* (X & 0b100) >> 2 */ 751 brw_OR(&func, Yp, t1, t2); 752 SWAP_XY_AND_XPYP(); 753 } 754} 755 756/** 757 * Emit code to compensate for the difference between MSAA and non-MSAA 758 * surfaces. 759 * 760 * This code modifies the X and Y coordinates according to the formula: 761 * 762 * (X', Y') = encode_msaa_4x(X, Y, S) 763 * 764 * (See brw_blorp_blit_program). 765 */ 766void 767brw_blorp_blit_program::encode_msaa(unsigned num_samples) 768{ 769 if (num_samples == 0) { 770 /* No translation necessary. */ 771 } else { 772 /* encode_msaa_4x(X, Y, S) = (X', Y') 773 * where X' = (X & ~0b1) << 1 | (S & 0b1) << 1 | (X & 0b1) 774 * Y' = (Y & ~0b1 ) << 1 | (S & 0b10) | (Y & 0b1) 775 */ 776 brw_AND(&func, t1, X, brw_imm_uw(0xfffe)); /* X & ~0b1 */ 777 if (!s_is_zero) { 778 brw_AND(&func, t2, S, brw_imm_uw(1)); /* S & 0b1 */ 779 brw_OR(&func, t1, t1, t2); /* (X & ~0b1) | (S & 0b1) */ 780 } 781 brw_SHL(&func, t1, t1, brw_imm_uw(1)); /* (X & ~0b1) << 1 782 | (S & 0b1) << 1 */ 783 brw_AND(&func, t2, X, brw_imm_uw(1)); /* X & 0b1 */ 784 brw_OR(&func, Xp, t1, t2); 785 brw_AND(&func, t1, Y, brw_imm_uw(0xfffe)); /* Y & ~0b1 */ 786 brw_SHL(&func, t1, t1, brw_imm_uw(1)); /* (Y & ~0b1) << 1 */ 787 if (!s_is_zero) { 788 brw_AND(&func, t2, S, brw_imm_uw(2)); /* S & 0b10 */ 789 brw_OR(&func, t1, t1, t2); /* (Y & ~0b1) << 1 | (S & 0b10) */ 790 } 791 brw_AND(&func, t2, Y, brw_imm_uw(1)); 792 brw_OR(&func, Yp, t1, t2); 793 SWAP_XY_AND_XPYP(); 794 } 795} 796 797/** 798 * Emit code to compensate for the difference between MSAA and non-MSAA 799 * surfaces. 800 * 801 * This code modifies the X and Y coordinates according to the formula: 802 * 803 * (X', Y', S) = decode_msaa(num_samples, X, Y) 804 * 805 * (See brw_blorp_blit_program). 806 */ 807void 808brw_blorp_blit_program::decode_msaa(unsigned num_samples) 809{ 810 if (num_samples == 0) { 811 /* No translation necessary. */ 812 s_is_zero = true; 813 } else { 814 /* decode_msaa_4x(X, Y) = (X', Y', S) 815 * where X' = (X & ~0b11) >> 1 | (X & 0b1) 816 * Y' = (Y & ~0b11) >> 1 | (Y & 0b1) 817 * S = (Y & 0b10) | (X & 0b10) >> 1 818 */ 819 brw_AND(&func, t1, X, brw_imm_uw(0xfffc)); /* X & ~0b11 */ 820 brw_SHR(&func, t1, t1, brw_imm_uw(1)); /* (X & ~0b11) >> 1 */ 821 brw_AND(&func, t2, X, brw_imm_uw(1)); /* X & 0b1 */ 822 brw_OR(&func, Xp, t1, t2); 823 brw_AND(&func, t1, Y, brw_imm_uw(0xfffc)); /* Y & ~0b11 */ 824 brw_SHR(&func, t1, t1, brw_imm_uw(1)); /* (Y & ~0b11) >> 1 */ 825 brw_AND(&func, t2, Y, brw_imm_uw(1)); /* Y & 0b1 */ 826 brw_OR(&func, Yp, t1, t2); 827 brw_AND(&func, t1, Y, brw_imm_uw(2)); /* Y & 0b10 */ 828 brw_AND(&func, t2, X, brw_imm_uw(2)); /* X & 0b10 */ 829 brw_SHR(&func, t2, t2, brw_imm_uw(1)); /* (X & 0b10) >> 1 */ 830 brw_OR(&func, S, t1, t2); 831 s_is_zero = false; 832 SWAP_XY_AND_XPYP(); 833 } 834} 835 836/** 837 * Emit code that kills pixels whose X and Y coordinates are outside the 838 * boundary of the rectangle defined by the push constants (dst_x0, dst_y0, 839 * dst_x1, dst_y1). 840 */ 841void 842brw_blorp_blit_program::kill_if_outside_dst_rect() 843{ 844 struct brw_reg f0 = brw_flag_reg(); 845 struct brw_reg g1 = retype(brw_vec1_grf(1, 7), BRW_REGISTER_TYPE_UW); 846 struct brw_reg null16 = vec16(retype(brw_null_reg(), BRW_REGISTER_TYPE_UW)); 847 848 brw_CMP(&func, null16, BRW_CONDITIONAL_GE, X, dst_x0); 849 brw_CMP(&func, null16, BRW_CONDITIONAL_GE, Y, dst_y0); 850 brw_CMP(&func, null16, BRW_CONDITIONAL_L, X, dst_x1); 851 brw_CMP(&func, null16, BRW_CONDITIONAL_L, Y, dst_y1); 852 853 brw_set_predicate_control(&func, BRW_PREDICATE_NONE); 854 brw_push_insn_state(&func); 855 brw_set_mask_control(&func, BRW_MASK_DISABLE); 856 brw_AND(&func, g1, f0, g1); 857 brw_pop_insn_state(&func); 858} 859 860/** 861 * Emit code to translate from destination (X, Y) coordinates to source (X, Y) 862 * coordinates. 863 */ 864void 865brw_blorp_blit_program::translate_dst_to_src() 866{ 867 brw_MUL(&func, Xp, X, x_transform.multiplier); 868 brw_MUL(&func, Yp, Y, y_transform.multiplier); 869 brw_ADD(&func, Xp, Xp, x_transform.offset); 870 brw_ADD(&func, Yp, Yp, y_transform.offset); 871 SWAP_XY_AND_XPYP(); 872} 873 874/** 875 * Emit code to transform the X and Y coordinates as needed for blending 876 * together the different samples in an MSAA texture. 877 */ 878void 879brw_blorp_blit_program::single_to_blend() 880{ 881 /* When looking up samples in an MSAA texture using the SAMPLE message, 882 * Gen6 requires the texture coordinates to be odd integers (so that they 883 * correspond to the center of a 2x2 block representing the four samples 884 * that maxe up a pixel). So we need to multiply our X and Y coordinates 885 * each by 2 and then add 1. 886 */ 887 brw_SHL(&func, t1, X, brw_imm_w(1)); 888 brw_SHL(&func, t2, Y, brw_imm_w(1)); 889 brw_ADD(&func, Xp, t1, brw_imm_w(1)); 890 brw_ADD(&func, Yp, t2, brw_imm_w(1)); 891 SWAP_XY_AND_XPYP(); 892} 893 894/** 895 * Emit code to look up a value in the texture using the SAMPLE message (which 896 * does blending of MSAA surfaces). 897 */ 898void 899brw_blorp_blit_program::sample() 900{ 901 texture_lookup(GEN5_SAMPLER_MESSAGE_SAMPLE, mrf_u_float, mrf_v_float); 902} 903 904/** 905 * Emit code to look up a value in the texture using the SAMPLE_LD message 906 * (which does a simple texel fetch). 907 */ 908void 909brw_blorp_blit_program::texel_fetch() 910{ 911 assert(s_is_zero); 912 texture_lookup(GEN5_SAMPLER_MESSAGE_SAMPLE_LD, 913 retype(mrf_u_float, BRW_REGISTER_TYPE_UD), 914 retype(mrf_v_float, BRW_REGISTER_TYPE_UD)); 915} 916 917void 918brw_blorp_blit_program::texture_lookup(GLuint msg_type, 919 struct brw_reg mrf_u, 920 struct brw_reg mrf_v) 921{ 922 /* Expand X and Y coordinates from 16 bits to 32 bits. */ 923 brw_MOV(&func, vec8(mrf_u), vec8(X)); 924 brw_set_compression_control(&func, BRW_COMPRESSION_2NDHALF); 925 brw_MOV(&func, offset(vec8(mrf_u), 1), suboffset(vec8(X), 8)); 926 brw_set_compression_control(&func, BRW_COMPRESSION_NONE); 927 brw_MOV(&func, vec8(mrf_v), vec8(Y)); 928 brw_set_compression_control(&func, BRW_COMPRESSION_2NDHALF); 929 brw_MOV(&func, offset(vec8(mrf_v), 1), suboffset(vec8(Y), 8)); 930 brw_set_compression_control(&func, BRW_COMPRESSION_NONE); 931 932 brw_SAMPLE(&func, 933 retype(Rdata, BRW_REGISTER_TYPE_UW) /* dest */, 934 base_mrf /* msg_reg_nr */, 935 vec8(mrf_u) /* src0 */, 936 BRW_BLORP_TEXTURE_BINDING_TABLE_INDEX, 937 0 /* sampler -- ignored for SAMPLE_LD message */, 938 WRITEMASK_XYZW, 939 msg_type, 940 8 /* response_length. TODO: should be smaller for non-RGBA formats? */, 941 4 /* msg_length */, 942 0 /* header_present */, 943 BRW_SAMPLER_SIMD_MODE_SIMD16, 944 BRW_SAMPLER_RETURN_FORMAT_FLOAT32); 945} 946 947#undef X 948#undef Y 949#undef U 950#undef V 951#undef S 952#undef SWAP_XY_AND_XPYP 953 954void 955brw_blorp_blit_program::render_target_write() 956{ 957 struct brw_reg mrf_rt_write = vec16(brw_message_reg(base_mrf)); 958 int mrf_offset = 0; 959 960 /* If we may have killed pixels, then we need to send R0 and R1 in a header 961 * so that the render target knows which pixels we killed. 962 */ 963 bool use_header = key->use_kill; 964 if (use_header) { 965 /* Copy R0/1 to MRF */ 966 brw_MOV(&func, retype(mrf_rt_write, BRW_REGISTER_TYPE_UD), 967 retype(R0, BRW_REGISTER_TYPE_UD)); 968 mrf_offset += 2; 969 } 970 971 /* Copy texture data to MRFs */ 972 for (int i = 0; i < 4; ++i) { 973 /* E.g. mov(16) m2.0<1>:f r2.0<8;8,1>:f { Align1, H1 } */ 974 brw_MOV(&func, offset(mrf_rt_write, mrf_offset), offset(vec8(Rdata), 2*i)); 975 mrf_offset += 2; 976 } 977 978 /* Now write to the render target and terminate the thread */ 979 brw_fb_WRITE(&func, 980 16 /* dispatch_width */, 981 base_mrf /* msg_reg_nr */, 982 mrf_rt_write /* src0 */, 983 BRW_BLORP_RENDERBUFFER_BINDING_TABLE_INDEX, 984 mrf_offset /* msg_length. TODO: Should be smaller for non-RGBA formats. */, 985 0 /* response_length */, 986 true /* eot */, 987 use_header); 988} 989 990 991void 992brw_blorp_coord_transform_params::setup(GLuint src0, GLuint dst0, GLuint dst1, 993 bool mirror) 994{ 995 if (!mirror) { 996 /* When not mirroring a coordinate (say, X), we need: 997 * x' - src_x0 = x - dst_x0 998 * Therefore: 999 * x' = 1*x + (src_x0 - dst_x0) 1000 */ 1001 multiplier = 1; 1002 offset = src0 - dst0; 1003 } else { 1004 /* When mirroring X we need: 1005 * x' - src_x0 = dst_x1 - x - 1 1006 * Therefore: 1007 * x' = -1*x + (src_x0 + dst_x1 - 1) 1008 */ 1009 multiplier = -1; 1010 offset = src0 + dst1 - 1; 1011 } 1012} 1013 1014 1015brw_blorp_blit_params::brw_blorp_blit_params(struct intel_mipmap_tree *src_mt, 1016 struct intel_mipmap_tree *dst_mt, 1017 GLuint src_x0, GLuint src_y0, 1018 GLuint dst_x0, GLuint dst_y0, 1019 GLuint dst_x1, GLuint dst_y1, 1020 bool mirror_x, bool mirror_y) 1021{ 1022 src.set(src_mt, 0, 0); 1023 dst.set(dst_mt, 0, 0); 1024 1025 use_wm_prog = true; 1026 memset(&wm_prog_key, 0, sizeof(wm_prog_key)); 1027 1028 if (src_mt->num_samples > 0 && dst_mt->num_samples > 0) { 1029 /* We are blitting from a multisample buffer to a multisample buffer, so 1030 * we must preserve samples within a pixel. This means we have to 1031 * configure the render target and texture surface states as 1032 * single-sampled, so that the WM program can access each sample 1033 * individually. 1034 */ 1035 src.num_samples = dst.num_samples = 0; 1036 } 1037 1038 /* The render path must be configured to use the same number of samples as 1039 * the destination buffer. 1040 */ 1041 num_samples = dst.num_samples; 1042 1043 GLenum base_format = _mesa_get_format_base_format(src_mt->format); 1044 if (base_format != GL_DEPTH_COMPONENT && /* TODO: what about depth/stencil? */ 1045 base_format != GL_STENCIL_INDEX && 1046 src_mt->num_samples > 0 && dst_mt->num_samples == 0) { 1047 /* We are downsampling a color buffer, so blend. */ 1048 wm_prog_key.blend = true; 1049 } 1050 1051 /* src_samples and dst_samples are the true sample counts */ 1052 wm_prog_key.src_samples = src_mt->num_samples; 1053 wm_prog_key.dst_samples = dst_mt->num_samples; 1054 1055 /* tex_samples and rt_samples are the sample counts that are set up in 1056 * SURFACE_STATE. 1057 */ 1058 wm_prog_key.tex_samples = src.num_samples; 1059 wm_prog_key.rt_samples = dst.num_samples; 1060 1061 wm_prog_key.src_tiled_w = src.map_stencil_as_y_tiled; 1062 wm_prog_key.dst_tiled_w = dst.map_stencil_as_y_tiled; 1063 x0 = wm_push_consts.dst_x0 = dst_x0; 1064 y0 = wm_push_consts.dst_y0 = dst_y0; 1065 x1 = wm_push_consts.dst_x1 = dst_x1; 1066 y1 = wm_push_consts.dst_y1 = dst_y1; 1067 wm_push_consts.x_transform.setup(src_x0, dst_x0, dst_x1, mirror_x); 1068 wm_push_consts.y_transform.setup(src_y0, dst_y0, dst_y1, mirror_y); 1069 1070 if (dst.num_samples == 0 && dst_mt->num_samples > 0) { 1071 /* We must expand the rectangle we send through the rendering pipeline, 1072 * to account for the fact that we are mapping the destination region as 1073 * single-sampled when it is in fact multisampled. We must also align 1074 * it to a multiple of the multisampling pattern, because the 1075 * differences between multisampled and single-sampled surface formats 1076 * will mean that pixels are scrambled within the multisampling pattern. 1077 * TODO: what if this makes the coordinates too large? 1078 */ 1079 x0 = (x0 * 2) & ~3; 1080 y0 = (y0 * 2) & ~3; 1081 x1 = ALIGN(x1 * 2, 4); 1082 y1 = ALIGN(y1 * 2, 4); 1083 wm_prog_key.use_kill = true; 1084 } 1085 1086 if (dst.map_stencil_as_y_tiled) { 1087 /* We must modify the rectangle we send through the rendering pipeline, 1088 * to account for the fact that we are mapping it as Y-tiled when it is 1089 * in fact W-tiled. Y tiles have dimensions 128x32 whereas W tiles have 1090 * dimensions 64x64. We must also align it to a multiple of the tile 1091 * size, because the differences between W and Y tiling formats will 1092 * mean that pixels are scrambled within the tile. 1093 * TODO: what if this makes the coordinates too large? 1094 */ 1095 x0 = (x0 * 2) & ~127; 1096 y0 = (y0 / 2) & ~31; 1097 x1 = ALIGN(x1 * 2, 128); 1098 y1 = ALIGN(y1 / 2, 32); 1099 wm_prog_key.use_kill = true; 1100 } 1101} 1102 1103uint32_t 1104brw_blorp_blit_params::get_wm_prog(struct brw_context *brw, 1105 brw_blorp_prog_data **prog_data) const 1106{ 1107 uint32_t prog_offset; 1108 if (!brw_search_cache(&brw->cache, BRW_BLORP_BLIT_PROG, 1109 &this->wm_prog_key, sizeof(this->wm_prog_key), 1110 &prog_offset, prog_data)) { 1111 brw_blorp_blit_program prog(brw, &this->wm_prog_key); 1112 GLuint program_size; 1113 const GLuint *program = prog.compile(brw, &program_size); 1114 brw_upload_cache(&brw->cache, BRW_BLORP_BLIT_PROG, 1115 &this->wm_prog_key, sizeof(this->wm_prog_key), 1116 program, program_size, 1117 &prog.prog_data, sizeof(prog.prog_data), 1118 &prog_offset, prog_data); 1119 } 1120 return prog_offset; 1121} 1122