1 /**************************************************************************
3 * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
30 #include "main/glheader.h"
31 #include "main/mtypes.h"
32 #include "main/imports.h"
33 #include "main/macros.h"
34 #include "main/colormac.h"
36 #include "tnl/t_context.h"
37 #include "tnl/t_vertex.h"
39 #include "intel_batchbuffer.h"
40 #include "intel_regions.h"
41 #include "intel_tris.h"
42 #include "intel_fbo.h"
45 #include "i915_context.h"
48 i915_render_prevalidate(struct intel_context *intel)
50 struct i915_context *i915 = i915_context(&intel->ctx);
52 i915ValidateFragmentProgram(i915);
56 i915_render_start(struct intel_context *intel)
58 intel_prepare_render(intel);
63 i915_reduced_primitive_state(struct intel_context *intel, GLenum rprim)
65 struct i915_context *i915 = i915_context(&intel->ctx);
66 GLuint st1 = i915->state.Stipple[I915_STPREG_ST1];
71 case GL_QUADS: /* from RASTERIZE(GL_QUADS) in t_dd_tritemp.h */
73 if (intel->ctx.Polygon.StippleFlag && intel->hw_stipple)
82 i915->intel.reduced_primitive = rprim;
84 if (st1 != i915->state.Stipple[I915_STPREG_ST1]) {
85 INTEL_FIREVERTICES(intel);
87 I915_STATECHANGE(i915, I915_UPLOAD_STIPPLE);
88 i915->state.Stipple[I915_STPREG_ST1] = st1;
93 /* Pull apart the vertex format registers and figure out how large a
94 * vertex is supposed to be.
97 i915_check_vertex_size(struct intel_context *intel, GLuint expected)
99 struct i915_context *i915 = i915_context(&intel->ctx);
100 int lis2 = i915->state.Ctx[I915_CTXREG_LIS2];
101 int lis4 = i915->state.Ctx[I915_CTXREG_LIS4];
104 switch (lis4 & S4_VFMT_XYZW_MASK) {
118 fprintf(stderr, "no xyzw specified\n");
122 if (lis4 & S4_VFMT_SPEC_FOG)
124 if (lis4 & S4_VFMT_COLOR)
126 if (lis4 & S4_VFMT_DEPTH_OFFSET)
128 if (lis4 & S4_VFMT_POINT_WIDTH)
130 if (lis4 & S4_VFMT_FOG_PARAM)
133 for (i = 0; i < 8; i++) {
134 switch (lis2 & S2_TEXCOORD_FMT0_MASK) {
147 case TEXCOORDFMT_2D_16:
150 case TEXCOORDFMT_4D_16:
153 case TEXCOORDFMT_NOT_PRESENT:
156 fprintf(stderr, "bad texcoord fmt %d\n", i);
159 lis2 >>= S2_TEXCOORD_FMT1_SHIFT;
163 fprintf(stderr, "vertex size mismatch %d/%d\n", sz, expected);
165 return sz == expected;
170 i915_emit_invarient_state(struct intel_context *intel)
176 OUT_BATCH(_3DSTATE_AA_CMD |
177 AA_LINE_ECAAR_WIDTH_ENABLE |
178 AA_LINE_ECAAR_WIDTH_1_0 |
179 AA_LINE_REGION_WIDTH_ENABLE | AA_LINE_REGION_WIDTH_1_0);
181 OUT_BATCH(_3DSTATE_DFLT_DIFFUSE_CMD);
184 OUT_BATCH(_3DSTATE_DFLT_SPEC_CMD);
187 OUT_BATCH(_3DSTATE_DFLT_Z_CMD);
190 /* Don't support texture crossbar yet */
191 OUT_BATCH(_3DSTATE_COORD_SET_BINDINGS |
196 CSB_TCB(4, 4) | CSB_TCB(5, 5) | CSB_TCB(6, 6) | CSB_TCB(7, 7));
198 /* Need to initialize this to zero.
200 OUT_BATCH(_3DSTATE_LOAD_STATE_IMMEDIATE_1 | I1_LOAD_S(3) | (0));
204 OUT_BATCH(_3DSTATE_SCISSOR_ENABLE_CMD | DISABLE_SCISSOR_RECT);
206 OUT_BATCH(_3DSTATE_SCISSOR_RECT_0_CMD);
210 OUT_BATCH(_3DSTATE_DEPTH_SUBRECT_DISABLE);
212 OUT_BATCH(_3DSTATE_LOAD_INDIRECT | 0); /* disable indirect state */
219 #define emit(intel, state, size ) \
220 intel_batchbuffer_data(intel, state, size, false)
223 get_dirty(struct i915_hw_state *state)
227 /* Workaround the multitex hang - if one texture unit state is
228 * modified, emit all texture units.
230 dirty = state->active & ~state->emitted;
231 if (dirty & I915_UPLOAD_TEX_ALL)
232 state->emitted &= ~I915_UPLOAD_TEX_ALL;
233 dirty = state->active & ~state->emitted;
239 get_state_size(struct i915_hw_state *state)
241 GLuint dirty = get_dirty(state);
245 if (dirty & I915_UPLOAD_INVARIENT)
248 if (dirty & I915_UPLOAD_RASTER_RULES)
249 sz += sizeof(state->RasterRules);
251 if (dirty & I915_UPLOAD_CTX)
252 sz += sizeof(state->Ctx);
254 if (dirty & I915_UPLOAD_BLEND)
255 sz += sizeof(state->Blend);
257 if (dirty & I915_UPLOAD_BUFFERS)
258 sz += sizeof(state->Buffer);
260 if (dirty & I915_UPLOAD_STIPPLE)
261 sz += sizeof(state->Stipple);
263 if (dirty & I915_UPLOAD_TEX_ALL) {
265 for (i = 0; i < I915_TEX_UNITS; i++)
266 if (dirty & I915_UPLOAD_TEX(i))
269 sz += (2 + nr * 3) * sizeof(GLuint) * 2;
272 if (dirty & I915_UPLOAD_CONSTANTS)
273 sz += state->ConstantSize * sizeof(GLuint);
275 if (dirty & I915_UPLOAD_PROGRAM)
276 sz += state->ProgramSize * sizeof(GLuint);
281 /* Push the state into the sarea and/or texture memory.
284 i915_emit_state(struct intel_context *intel)
286 struct i915_context *i915 = i915_context(&intel->ctx);
287 struct i915_hw_state *state = &i915->state;
288 int i, count, aper_count;
290 drm_intel_bo *aper_array[3 + I915_TEX_UNITS];
291 GET_CURRENT_CONTEXT(ctx);
294 /* We don't hold the lock at this point, so want to make sure that
295 * there won't be a buffer wrap between the state emits and the primitive
298 * It might be better to talk about explicit places where
299 * scheduling is allowed, rather than assume that it is whenever a
300 * batchbuffer fills up.
302 intel_batchbuffer_require_space(intel,
303 get_state_size(state) + INTEL_PRIM_EMIT_SIZE,
307 if (intel->batch.bo == NULL) {
308 _mesa_error(ctx, GL_OUT_OF_MEMORY, "i915 emit state");
312 dirty = get_dirty(state);
314 aper_array[aper_count++] = intel->batch.bo;
315 if (dirty & I915_UPLOAD_BUFFERS) {
316 aper_array[aper_count++] = state->draw_region->buffer;
317 if (state->depth_region)
318 aper_array[aper_count++] = state->depth_region->buffer;
321 if (dirty & I915_UPLOAD_TEX_ALL) {
322 for (i = 0; i < I915_TEX_UNITS; i++) {
323 if (dirty & I915_UPLOAD_TEX(i)) {
324 if (state->tex_buffer[i]) {
325 aper_array[aper_count++] = state->tex_buffer[i];
331 if (dri_bufmgr_check_aperture_space(aper_array, aper_count)) {
334 intel_batchbuffer_flush(intel);
337 _mesa_error(ctx, GL_OUT_OF_MEMORY, "i915 emit state");
342 /* work out list of buffers to emit */
344 /* Do this here as we may have flushed the batchbuffer above,
345 * causing more state to be dirty!
347 dirty = get_dirty(state);
348 state->emitted |= dirty;
349 assert(get_dirty(state) == 0);
351 if (INTEL_DEBUG & DEBUG_STATE)
352 fprintf(stderr, "%s dirty: %x\n", __FUNCTION__, dirty);
354 if (dirty & I915_UPLOAD_INVARIENT) {
355 if (INTEL_DEBUG & DEBUG_STATE)
356 fprintf(stderr, "I915_UPLOAD_INVARIENT:\n");
357 i915_emit_invarient_state(intel);
360 if (dirty & I915_UPLOAD_RASTER_RULES) {
361 if (INTEL_DEBUG & DEBUG_STATE)
362 fprintf(stderr, "I915_UPLOAD_RASTER_RULES:\n");
363 emit(intel, state->RasterRules, sizeof(state->RasterRules));
366 if (dirty & I915_UPLOAD_CTX) {
367 if (INTEL_DEBUG & DEBUG_STATE)
368 fprintf(stderr, "I915_UPLOAD_CTX:\n");
370 emit(intel, state->Ctx, sizeof(state->Ctx));
373 if (dirty & I915_UPLOAD_BLEND) {
374 if (INTEL_DEBUG & DEBUG_STATE)
375 fprintf(stderr, "I915_UPLOAD_BLEND:\n");
377 emit(intel, state->Blend, sizeof(state->Blend));
380 if (dirty & I915_UPLOAD_BUFFERS) {
383 if (INTEL_DEBUG & DEBUG_STATE)
384 fprintf(stderr, "I915_UPLOAD_BUFFERS:\n");
387 if (state->Buffer[I915_DESTREG_DRAWRECT0] != MI_NOOP)
389 if (state->depth_region)
393 OUT_BATCH(state->Buffer[I915_DESTREG_CBUFADDR0]);
394 OUT_BATCH(state->Buffer[I915_DESTREG_CBUFADDR1]);
395 OUT_RELOC(state->draw_region->buffer,
396 I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, 0);
398 if (state->depth_region) {
399 OUT_BATCH(state->Buffer[I915_DESTREG_DBUFADDR0]);
400 OUT_BATCH(state->Buffer[I915_DESTREG_DBUFADDR1]);
401 OUT_RELOC(state->depth_region->buffer,
402 I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, 0);
405 OUT_BATCH(state->Buffer[I915_DESTREG_DV0]);
406 OUT_BATCH(state->Buffer[I915_DESTREG_DV1]);
407 OUT_BATCH(state->Buffer[I915_DESTREG_SENABLE]);
408 OUT_BATCH(state->Buffer[I915_DESTREG_SR0]);
409 OUT_BATCH(state->Buffer[I915_DESTREG_SR1]);
410 OUT_BATCH(state->Buffer[I915_DESTREG_SR2]);
412 if (state->Buffer[I915_DESTREG_DRAWRECT0] != MI_NOOP)
413 OUT_BATCH(state->Buffer[I915_DESTREG_DRAWRECT0]);
414 OUT_BATCH(state->Buffer[I915_DESTREG_DRAWRECT1]);
415 OUT_BATCH(state->Buffer[I915_DESTREG_DRAWRECT2]);
416 OUT_BATCH(state->Buffer[I915_DESTREG_DRAWRECT3]);
417 OUT_BATCH(state->Buffer[I915_DESTREG_DRAWRECT4]);
418 OUT_BATCH(state->Buffer[I915_DESTREG_DRAWRECT5]);
423 if (dirty & I915_UPLOAD_STIPPLE) {
424 if (INTEL_DEBUG & DEBUG_STATE)
425 fprintf(stderr, "I915_UPLOAD_STIPPLE:\n");
426 emit(intel, state->Stipple, sizeof(state->Stipple));
429 /* Combine all the dirty texture state into a single command to
430 * avoid lockups on I915 hardware.
432 if (dirty & I915_UPLOAD_TEX_ALL) {
436 for (i = 0; i < I915_TEX_UNITS; i++)
437 if (dirty & I915_UPLOAD_TEX(i))
440 BEGIN_BATCH(2 + nr * 3);
441 OUT_BATCH(_3DSTATE_MAP_STATE | (3 * nr));
442 OUT_BATCH((dirty & I915_UPLOAD_TEX_ALL) >> I915_UPLOAD_TEX_0_SHIFT);
443 for (i = 0; i < I915_TEX_UNITS; i++)
444 if (dirty & I915_UPLOAD_TEX(i)) {
445 OUT_RELOC(state->tex_buffer[i],
446 I915_GEM_DOMAIN_SAMPLER, 0,
447 state->tex_offset[i]);
449 OUT_BATCH(state->Tex[i][I915_TEXREG_MS3]);
450 OUT_BATCH(state->Tex[i][I915_TEXREG_MS4]);
454 unwind = intel->batch.used;
455 BEGIN_BATCH(2 + nr * 3);
456 OUT_BATCH(_3DSTATE_SAMPLER_STATE | (3 * nr));
457 OUT_BATCH((dirty & I915_UPLOAD_TEX_ALL) >> I915_UPLOAD_TEX_0_SHIFT);
458 for (i = 0; i < I915_TEX_UNITS; i++)
459 if (dirty & I915_UPLOAD_TEX(i)) {
460 OUT_BATCH(state->Tex[i][I915_TEXREG_SS2]);
461 OUT_BATCH(state->Tex[i][I915_TEXREG_SS3]);
462 OUT_BATCH(state->Tex[i][I915_TEXREG_SS4]);
465 if (i915->last_sampler &&
466 memcmp(intel->batch.map + i915->last_sampler,
467 intel->batch.map + unwind,
468 (2 + nr*3)*sizeof(int)) == 0)
469 intel->batch.used = unwind;
471 i915->last_sampler = unwind;
474 if (dirty & I915_UPLOAD_CONSTANTS) {
475 if (INTEL_DEBUG & DEBUG_STATE)
476 fprintf(stderr, "I915_UPLOAD_CONSTANTS:\n");
477 emit(intel, state->Constant, state->ConstantSize * sizeof(GLuint));
480 if (dirty & I915_UPLOAD_PROGRAM) {
481 if (state->ProgramSize) {
482 if (INTEL_DEBUG & DEBUG_STATE)
483 fprintf(stderr, "I915_UPLOAD_PROGRAM:\n");
485 assert((state->Program[0] & 0x1ff) + 2 == state->ProgramSize);
487 emit(intel, state->Program, state->ProgramSize * sizeof(GLuint));
488 if (INTEL_DEBUG & DEBUG_STATE)
489 i915_disassemble_program(state->Program, state->ProgramSize);
493 assert(get_dirty(state) == 0);
497 i915_destroy_context(struct intel_context *intel)
500 struct i915_context *i915 = i915_context(&intel->ctx);
502 intel_region_release(&i915->state.draw_region);
503 intel_region_release(&i915->state.depth_region);
505 for (i = 0; i < I915_TEX_UNITS; i++) {
506 if (i915->state.tex_buffer[i] != NULL) {
507 drm_intel_bo_unreference(i915->state.tex_buffer[i]);
508 i915->state.tex_buffer[i] = NULL;
512 _tnl_free_vertices(&intel->ctx);
516 i915_set_buf_info_for_region(uint32_t *state, struct intel_region *region,
519 state[0] = _3DSTATE_BUF_INFO_CMD;
520 state[1] = buffer_id;
522 if (region != NULL) {
523 state[1] |= BUF_3D_PITCH(region->pitch * region->cpp);
525 if (region->tiling != I915_TILING_NONE) {
526 state[1] |= BUF_3D_TILED_SURFACE;
527 if (region->tiling == I915_TILING_Y)
528 state[1] |= BUF_3D_TILE_WALK_Y;
533 static uint32_t i915_render_target_format_for_mesa_format[MESA_FORMAT_COUNT] =
535 [MESA_FORMAT_ARGB8888] = DV_PF_8888,
536 [MESA_FORMAT_XRGB8888] = DV_PF_8888,
537 [MESA_FORMAT_RGB565] = DV_PF_565 | DITHER_FULL_ALWAYS,
538 [MESA_FORMAT_ARGB1555] = DV_PF_1555 | DITHER_FULL_ALWAYS,
539 [MESA_FORMAT_ARGB4444] = DV_PF_4444 | DITHER_FULL_ALWAYS,
543 i915_render_target_supported(gl_format format)
545 if (format == MESA_FORMAT_S8_Z24 ||
546 format == MESA_FORMAT_X8_Z24 ||
547 format == MESA_FORMAT_Z16) {
551 return i915_render_target_format_for_mesa_format[format] != 0;
555 i915_set_draw_region(struct intel_context *intel,
556 struct intel_region *color_regions[],
557 struct intel_region *depth_region,
560 struct i915_context *i915 = i915_context(&intel->ctx);
561 struct gl_context *ctx = &intel->ctx;
562 struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[0];
563 struct intel_renderbuffer *irb = intel_renderbuffer(rb);
565 struct i915_hw_state *state = &i915->state;
566 uint32_t draw_x, draw_y, draw_offset;
568 if (state->draw_region != color_regions[0]) {
569 intel_region_release(&state->draw_region);
570 intel_region_reference(&state->draw_region, color_regions[0]);
572 if (state->depth_region != depth_region) {
573 intel_region_release(&state->depth_region);
574 intel_region_reference(&state->depth_region, depth_region);
578 * Set stride/cpp values
580 i915_set_buf_info_for_region(&state->Buffer[I915_DESTREG_CBUFADDR0],
581 color_regions[0], BUF_3D_ID_COLOR_BACK);
583 i915_set_buf_info_for_region(&state->Buffer[I915_DESTREG_DBUFADDR0],
584 depth_region, BUF_3D_ID_DEPTH);
587 * Compute/set I915_DESTREG_DV1 value
589 value = (DSTORG_HORT_BIAS(0x8) | /* .5 */
590 DSTORG_VERT_BIAS(0x8) | /* .5 */
591 LOD_PRECLAMP_OGL | TEX_DEFAULT_COLOR_OGL);
593 value |= i915_render_target_format_for_mesa_format[irb->Base.Format];
596 /* This isn't quite safe, thus being hidden behind an option. When changing
597 * the value of this bit, the pipeline needs to be MI_FLUSHed. And it
598 * can only be set when a depth buffer is already defined.
600 if (intel->is_945 && intel->use_early_z &&
601 depth_region->tiling != I915_TILING_NONE)
602 value |= CLASSIC_EARLY_DEPTH;
604 if (depth_region && depth_region->cpp == 4) {
605 value |= DEPTH_FRMT_24_FIXED_8_OTHER;
608 value |= DEPTH_FRMT_16_FIXED;
610 state->Buffer[I915_DESTREG_DV1] = value;
612 /* We set up the drawing rectangle to be offset into the color
613 * region's location in the miptree. If it doesn't match with
614 * depth's offsets, we can't render to it.
616 * (Well, not actually true -- the hw grew a bit to let depth's
617 * offset get forced to 0,0. We may want to use that if people are
618 * hitting that case. Also, some configurations may be supportable
619 * by tweaking the start offset of the buffers around, which we
620 * can't do in general due to tiling)
622 FALLBACK(intel, I915_FALLBACK_DRAW_OFFSET,
623 (depth_region && color_regions[0]) &&
624 (depth_region->draw_x != color_regions[0]->draw_x ||
625 depth_region->draw_y != color_regions[0]->draw_y));
627 if (color_regions[0]) {
628 draw_x = color_regions[0]->draw_x;
629 draw_y = color_regions[0]->draw_y;
630 } else if (depth_region) {
631 draw_x = depth_region->draw_x;
632 draw_y = depth_region->draw_y;
638 draw_offset = (draw_y << 16) | draw_x;
640 /* When changing drawing rectangle offset, an MI_FLUSH is first required. */
641 if (draw_offset != i915->last_draw_offset) {
642 FALLBACK(intel, I915_FALLBACK_DRAW_OFFSET,
643 (ctx->DrawBuffer->Width + draw_x > 2048) ||
644 (ctx->DrawBuffer->Height + draw_y > 2048));
646 state->Buffer[I915_DESTREG_DRAWRECT0] = MI_FLUSH | INHIBIT_FLUSH_RENDER_CACHE;
647 i915->last_draw_offset = draw_offset;
649 state->Buffer[I915_DESTREG_DRAWRECT0] = MI_NOOP;
651 state->Buffer[I915_DESTREG_DRAWRECT1] = _3DSTATE_DRAWRECT_INFO;
652 state->Buffer[I915_DESTREG_DRAWRECT2] = 0;
653 state->Buffer[I915_DESTREG_DRAWRECT3] = draw_offset;
654 state->Buffer[I915_DESTREG_DRAWRECT4] =
655 ((ctx->DrawBuffer->Width + draw_x - 1) & 0xffff) |
656 ((ctx->DrawBuffer->Height + draw_y - 1) << 16);
657 state->Buffer[I915_DESTREG_DRAWRECT5] = draw_offset;
659 I915_STATECHANGE(i915, I915_UPLOAD_BUFFERS);
665 i915_new_batch(struct intel_context *intel)
667 struct i915_context *i915 = i915_context(&intel->ctx);
669 /* Mark all state as needing to be emitted when starting a new batchbuffer.
670 * Using hardware contexts would be an alternative, but they have some
671 * difficulties associated with them (physical address requirements).
673 i915->state.emitted = 0;
674 i915->last_draw_offset = 0;
675 i915->last_sampler = 0;
677 i915->current_vb_bo = NULL;
678 i915->current_vertex_size = 0;
682 i915_assert_not_dirty( struct intel_context *intel )
684 struct i915_context *i915 = i915_context(&intel->ctx);
685 GLuint dirty = get_dirty(&i915->state);
690 /** Return false; i915 does not support HiZ. */
692 i915_is_hiz_depth_format(struct intel_context *intel,
699 i915InitVtbl(struct i915_context *i915)
701 i915->intel.vtbl.check_vertex_size = i915_check_vertex_size;
702 i915->intel.vtbl.destroy = i915_destroy_context;
703 i915->intel.vtbl.emit_state = i915_emit_state;
704 i915->intel.vtbl.new_batch = i915_new_batch;
705 i915->intel.vtbl.reduced_primitive_state = i915_reduced_primitive_state;
706 i915->intel.vtbl.render_start = i915_render_start;
707 i915->intel.vtbl.render_prevalidate = i915_render_prevalidate;
708 i915->intel.vtbl.set_draw_region = i915_set_draw_region;
709 i915->intel.vtbl.update_texture_state = i915UpdateTextureState;
710 i915->intel.vtbl.assert_not_dirty = i915_assert_not_dirty;
711 i915->intel.vtbl.finish_batch = intel_finish_vb;
712 i915->intel.vtbl.render_target_supported = i915_render_target_supported;
713 i915->intel.vtbl.is_hiz_depth_format = i915_is_hiz_depth_format;