2 * Copyright © 2011 Intel Corporation
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:
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
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
23 #include "main/mtypes.h"
24 #include "main/samplerobj.h"
25 #include "program/prog_parameter.h"
27 #include "intel_mipmap_tree.h"
28 #include "intel_batchbuffer.h"
29 #include "intel_tex.h"
30 #include "intel_fbo.h"
31 #include "intel_buffer_objects.h"
33 #include "brw_context.h"
34 #include "brw_state.h"
35 #include "brw_defines.h"
39 * Convert an swizzle enumeration (i.e. SWIZZLE_X) to one of the Gen7.5+
40 * "Shader Channel Select" enumerations (i.e. HSW_SCS_RED)
43 swizzle_to_scs(GLenum swizzle)
60 assert(!"Should not get here: invalid swizzle mode");
65 gen7_surface_tiling_mode(uint32_t tiling)
69 return GEN7_SURFACE_TILING_X;
71 return GEN7_SURFACE_TILING_Y;
73 return GEN7_SURFACE_TILING_NONE;
79 gen7_surface_msaa_bits(unsigned num_samples, enum intel_msaa_layout layout)
84 ss4 |= GEN7_SURFACE_MULTISAMPLECOUNT_8;
85 else if (num_samples > 1)
86 ss4 |= GEN7_SURFACE_MULTISAMPLECOUNT_4;
88 ss4 |= GEN7_SURFACE_MULTISAMPLECOUNT_1;
90 if (layout == INTEL_MSAA_LAYOUT_IMS)
91 ss4 |= GEN7_SURFACE_MSFMT_DEPTH_STENCIL;
93 ss4 |= GEN7_SURFACE_MSFMT_MSS;
100 gen7_set_surface_mcs_info(struct brw_context *brw,
102 uint32_t surf_offset,
103 const struct intel_mipmap_tree *mcs_mt,
104 bool is_render_target)
106 /* From the Ivy Bridge PRM, Vol4 Part1 p76, "MCS Base Address":
108 * "The MCS surface must be stored as Tile Y."
110 assert(mcs_mt->region->tiling == I915_TILING_Y);
112 /* Compute the pitch in units of tiles. To do this we need to divide the
113 * pitch in bytes by 128, since a single Y-tile is 128 bytes wide.
115 unsigned pitch_bytes = mcs_mt->region->pitch * mcs_mt->cpp;
116 unsigned pitch_tiles = pitch_bytes / 128;
118 /* The upper 20 bits of surface state DWORD 6 are the upper 20 bits of the
119 * GPU address of the MCS buffer; the lower 12 bits contain other control
120 * information. Since buffer addresses are always on 4k boundaries (and
121 * thus have their lower 12 bits zero), we can use an ordinary reloc to do
122 * the necessary address translation.
124 assert ((mcs_mt->region->bo->offset & 0xfff) == 0);
126 surf[6] = GEN7_SURFACE_MCS_ENABLE |
127 SET_FIELD(pitch_tiles - 1, GEN7_SURFACE_MCS_PITCH) |
128 mcs_mt->region->bo->offset;
130 drm_intel_bo_emit_reloc(brw->intel.batch.bo,
134 is_render_target ? I915_GEM_DOMAIN_RENDER
135 : I915_GEM_DOMAIN_SAMPLER,
136 is_render_target ? I915_GEM_DOMAIN_RENDER : 0);
141 gen7_check_surface_setup(uint32_t *surf, bool is_render_target)
143 unsigned num_multisamples = surf[4] & INTEL_MASK(5, 3);
144 unsigned multisampled_surface_storage_format = surf[4] & (1 << 6);
145 unsigned surface_array_spacing = surf[0] & (1 << 10);
146 bool is_multisampled = num_multisamples != GEN7_SURFACE_MULTISAMPLECOUNT_1;
148 (void) surface_array_spacing;
150 /* From the Graphics BSpec: vol5c Shared Functions [SNB+] > State >
151 * SURFACE_STATE > SURFACE_STATE for most messages [DevIVB]: Surface Array
154 * If Multisampled Surface Storage Format is MSFMT_MSS and Number of
155 * Multisamples is not MULTISAMPLECOUNT_1, this field must be set to
158 if (multisampled_surface_storage_format == GEN7_SURFACE_MSFMT_MSS
160 assert(surface_array_spacing == GEN7_SURFACE_ARYSPC_LOD0);
162 /* From the Graphics BSpec: vol5c Shared Functions [SNB+] > State >
163 * SURFACE_STATE > SURFACE_STATE for most messages [DevIVB]: Multisampled
164 * Surface Storage Format:
166 * All multisampled render target surfaces must have this field set to
171 * This field is ignored if Number of Multisamples is MULTISAMPLECOUNT_1.
173 if (is_render_target && is_multisampled) {
174 assert(multisampled_surface_storage_format == GEN7_SURFACE_MSFMT_MSS);
177 /* From the Graphics BSpec: vol5c Shared Functions [SNB+] > State >
178 * SURFACE_STATE > SURFACE_STATE for most messages [DevIVB]: Multisampled
179 * Surface Storage Format:
181 * If the surface’s Number of Multisamples is MULTISAMPLECOUNT_8, Width
182 * is >= 8192 (meaning the actual surface width is >= 8193 pixels), this
183 * field must be set to MSFMT_MSS.
185 uint32_t width = GET_FIELD(surf[2], GEN7_SURFACE_WIDTH) + 1;
186 if (num_multisamples == GEN7_SURFACE_MULTISAMPLECOUNT_8 && width >= 8193) {
187 assert(multisampled_surface_storage_format == GEN7_SURFACE_MSFMT_MSS);
190 /* From the Graphics BSpec: vol5c Shared Functions [SNB+] > State >
191 * SURFACE_STATE > SURFACE_STATE for most messages [DevIVB]: Multisampled
192 * Surface Storage Format:
194 * If the surface’s Number of Multisamples is MULTISAMPLECOUNT_8,
195 * ((Depth+1) * (Height+1)) is > 4,194,304, OR if the surface’s Number of
196 * Multisamples is MULTISAMPLECOUNT_4, ((Depth+1) * (Height+1)) is >
197 * 8,388,608, this field must be set to MSFMT_DEPTH_STENCIL.This field
198 * must be set to MSFMT_DEPTH_STENCIL if Surface Format is one of the
199 * following: I24X8_UNORM, L24X8_UNORM, A24X8_UNORM, or
200 * R24_UNORM_X8_TYPELESS.
204 * This field is ignored if Number of Multisamples is MULTISAMPLECOUNT_1.
206 uint32_t depth = GET_FIELD(surf[3], BRW_SURFACE_DEPTH) + 1;
207 uint32_t height = GET_FIELD(surf[2], GEN7_SURFACE_HEIGHT) + 1;
208 if (num_multisamples == GEN7_SURFACE_MULTISAMPLECOUNT_8 &&
209 depth * height > 4194304) {
210 assert(multisampled_surface_storage_format ==
211 GEN7_SURFACE_MSFMT_DEPTH_STENCIL);
213 if (num_multisamples == GEN7_SURFACE_MULTISAMPLECOUNT_4 &&
214 depth * height > 8388608) {
215 assert(multisampled_surface_storage_format ==
216 GEN7_SURFACE_MSFMT_DEPTH_STENCIL);
218 if (is_multisampled) {
219 switch (GET_FIELD(surf[0], BRW_SURFACE_FORMAT)) {
220 case BRW_SURFACEFORMAT_I24X8_UNORM:
221 case BRW_SURFACEFORMAT_L24X8_UNORM:
222 case BRW_SURFACEFORMAT_A24X8_UNORM:
223 case BRW_SURFACEFORMAT_R24_UNORM_X8_TYPELESS:
224 assert(multisampled_surface_storage_format ==
225 GEN7_SURFACE_MSFMT_DEPTH_STENCIL);
232 gen7_update_buffer_texture_surface(struct gl_context *ctx,
234 uint32_t *binding_table,
237 struct brw_context *brw = brw_context(ctx);
238 struct intel_context *intel = &brw->intel;
239 struct gl_texture_object *tObj = ctx->Texture.Unit[unit]._Current;
240 struct intel_buffer_object *intel_obj =
241 intel_buffer_object(tObj->BufferObject);
242 drm_intel_bo *bo = intel_obj ? intel_obj->buffer : NULL;
243 gl_format format = tObj->_BufferObjectFormat;
245 uint32_t *surf = brw_state_batch(brw, AUB_TRACE_SURFACE_STATE,
246 8 * 4, 32, &binding_table[surf_index]);
247 memset(surf, 0, 8 * 4);
249 uint32_t surface_format = brw_format_for_mesa_format(format);
250 if (surface_format == 0 && format != MESA_FORMAT_RGBA_FLOAT32) {
251 _mesa_problem(NULL, "bad format %s for texture buffer\n",
252 _mesa_get_format_name(format));
255 surf[0] = BRW_SURFACE_BUFFER << BRW_SURFACE_TYPE_SHIFT |
256 surface_format << BRW_SURFACE_FORMAT_SHIFT |
257 BRW_SURFACE_RC_READ_WRITE;
260 surf[1] = bo->offset; /* reloc */
262 /* Emit relocation to surface contents. Section 5.1.1 of the gen4
263 * bspec ("Data Cache") says that the data cache does not exist as
264 * a separate cache and is just the sampler cache.
266 drm_intel_bo_emit_reloc(intel->batch.bo,
267 binding_table[surf_index] + 4,
269 I915_GEM_DOMAIN_SAMPLER, 0);
271 int texel_size = _mesa_get_format_bytes(format);
272 int w = intel_obj->Base.Size / texel_size;
273 surf[2] = SET_FIELD(w & 0x7f, GEN7_SURFACE_WIDTH) | /* bits 6:0 of size */
274 SET_FIELD((w >> 7) & 0x1fff, GEN7_SURFACE_HEIGHT); /* 19:7 */
275 surf[3] = SET_FIELD((w >> 20) & 0x7f, BRW_SURFACE_DEPTH) | /* bits 26:20 */
279 gen7_check_surface_setup(surf, false /* is_render_target */);
283 gen7_update_texture_surface(struct gl_context *ctx,
285 uint32_t *binding_table,
288 struct brw_context *brw = brw_context(ctx);
289 struct intel_context *intel = &brw->intel;
290 struct gl_texture_object *tObj = ctx->Texture.Unit[unit]._Current;
291 struct intel_texture_object *intelObj = intel_texture_object(tObj);
292 struct intel_mipmap_tree *mt = intelObj->mt;
293 struct gl_texture_image *firstImage = tObj->Image[0][tObj->BaseLevel];
294 struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit);
295 int width, height, depth;
297 if (tObj->Target == GL_TEXTURE_BUFFER) {
298 gen7_update_buffer_texture_surface(ctx, unit, binding_table, surf_index);
302 /* We don't support MSAA for textures. */
303 assert(!mt->array_spacing_lod0);
304 assert(mt->num_samples <= 1);
306 intel_miptree_get_dimensions_for_image(firstImage, &width, &height, &depth);
308 uint32_t *surf = brw_state_batch(brw, AUB_TRACE_SURFACE_STATE,
309 8 * 4, 32, &binding_table[surf_index]);
310 memset(surf, 0, 8 * 4);
312 uint32_t tex_format = translate_tex_format(mt->format,
313 firstImage->InternalFormat,
315 sampler->sRGBDecode);
317 surf[0] = translate_tex_target(tObj->Target) << BRW_SURFACE_TYPE_SHIFT |
318 tex_format << BRW_SURFACE_FORMAT_SHIFT |
319 gen7_surface_tiling_mode(mt->region->tiling) |
320 BRW_SURFACE_CUBEFACE_ENABLES;
322 if (mt->align_h == 4)
323 surf[0] |= GEN7_SURFACE_VALIGN_4;
324 if (mt->align_w == 8)
325 surf[0] |= GEN7_SURFACE_HALIGN_8;
327 if (depth > 1 && tObj->Target != GL_TEXTURE_3D)
328 surf[0] |= GEN7_SURFACE_IS_ARRAY;
330 surf[1] = mt->region->bo->offset + mt->offset; /* reloc */
332 surf[2] = SET_FIELD(width - 1, GEN7_SURFACE_WIDTH) |
333 SET_FIELD(height - 1, GEN7_SURFACE_HEIGHT);
334 surf[3] = SET_FIELD(depth - 1, BRW_SURFACE_DEPTH) |
335 ((intelObj->mt->region->pitch * intelObj->mt->cpp) - 1);
337 surf[5] = intelObj->_MaxLevel - tObj->BaseLevel; /* mip count */
339 if (intel->is_haswell) {
340 /* Handling GL_ALPHA as a surface format override breaks 1.30+ style
341 * texturing functions that return a float, as our code generation always
342 * selects the .x channel (which would always be 0).
344 const bool alpha_depth = tObj->DepthMode == GL_ALPHA &&
345 (firstImage->_BaseFormat == GL_DEPTH_COMPONENT ||
346 firstImage->_BaseFormat == GL_DEPTH_STENCIL);
349 unlikely(alpha_depth) ? SWIZZLE_XYZW : brw_get_texture_swizzle(tObj);
353 SET_FIELD(swizzle_to_scs(GET_SWZ(swizzle, 0)), GEN7_SURFACE_SCS_R) |
354 SET_FIELD(swizzle_to_scs(GET_SWZ(swizzle, 1)), GEN7_SURFACE_SCS_G) |
355 SET_FIELD(swizzle_to_scs(GET_SWZ(swizzle, 2)), GEN7_SURFACE_SCS_B) |
356 SET_FIELD(swizzle_to_scs(GET_SWZ(swizzle, 3)), GEN7_SURFACE_SCS_A);
359 /* Emit relocation to surface contents */
360 drm_intel_bo_emit_reloc(brw->intel.batch.bo,
361 binding_table[surf_index] + 4,
362 intelObj->mt->region->bo, intelObj->mt->offset,
363 I915_GEM_DOMAIN_SAMPLER, 0);
365 gen7_check_surface_setup(surf, false /* is_render_target */);
369 * Create the constant buffer surface. Vertex/fragment shader constants will
370 * be read from this buffer with Data Port Read instructions/messages.
373 gen7_create_constant_surface(struct brw_context *brw,
377 uint32_t *out_offset)
379 struct intel_context *intel = &brw->intel;
380 const GLint w = width - 1;
382 uint32_t *surf = brw_state_batch(brw, AUB_TRACE_SURFACE_STATE,
383 8 * 4, 32, out_offset);
384 memset(surf, 0, 8 * 4);
386 surf[0] = BRW_SURFACE_BUFFER << BRW_SURFACE_TYPE_SHIFT |
387 BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_SURFACE_FORMAT_SHIFT |
388 BRW_SURFACE_RC_READ_WRITE;
391 surf[1] = bo->offset + offset; /* reloc */
393 surf[2] = SET_FIELD(w & 0x7f, GEN7_SURFACE_WIDTH) |
394 SET_FIELD((w >> 7) & 0x1fff, GEN7_SURFACE_HEIGHT);
395 surf[3] = SET_FIELD((w >> 20) & 0x7f, BRW_SURFACE_DEPTH) |
396 (16 - 1); /* stride between samples */
398 if (intel->is_haswell) {
399 surf[7] = SET_FIELD(HSW_SCS_RED, GEN7_SURFACE_SCS_R) |
400 SET_FIELD(HSW_SCS_GREEN, GEN7_SURFACE_SCS_G) |
401 SET_FIELD(HSW_SCS_BLUE, GEN7_SURFACE_SCS_B) |
402 SET_FIELD(HSW_SCS_ALPHA, GEN7_SURFACE_SCS_A);
405 /* Emit relocation to surface contents. Section 5.1.1 of the gen4
406 * bspec ("Data Cache") says that the data cache does not exist as
407 * a separate cache and is just the sampler cache.
409 drm_intel_bo_emit_reloc(intel->batch.bo,
412 I915_GEM_DOMAIN_SAMPLER, 0);
414 gen7_check_surface_setup(surf, false /* is_render_target */);
418 gen7_update_null_renderbuffer_surface(struct brw_context *brw, unsigned unit)
420 /* From the Ivy bridge PRM, Vol4 Part1 p62 (Surface Type: Programming
423 * A null surface is used in instances where an actual surface is not
424 * bound. When a write message is generated to a null surface, no
425 * actual surface is written to. When a read message (including any
426 * sampling engine message) is generated to a null surface, the result
427 * is all zeros. Note that a null surface type is allowed to be used
428 * with all messages, even if it is not specificially indicated as
429 * supported. All of the remaining fields in surface state are ignored
430 * for null surfaces, with the following exceptions: Width, Height,
431 * Depth, LOD, and Render Target View Extent fields must match the
432 * depth buffer’s corresponding state for all render target surfaces,
435 struct intel_context *intel = &brw->intel;
436 struct gl_context *ctx = &intel->ctx;
439 const struct gl_framebuffer *fb = ctx->DrawBuffer;
441 uint32_t *surf = brw_state_batch(brw, AUB_TRACE_SURFACE_STATE,
442 8 * 4, 32, &brw->wm.surf_offset[unit]);
443 memset(surf, 0, 8 * 4);
445 /* From the Ivybridge PRM, Volume 4, Part 1, page 65,
446 * Tiled Surface: Programming Notes:
447 * "If Surface Type is SURFTYPE_NULL, this field must be TRUE."
449 surf[0] = BRW_SURFACE_NULL << BRW_SURFACE_TYPE_SHIFT |
450 BRW_SURFACEFORMAT_B8G8R8A8_UNORM << BRW_SURFACE_FORMAT_SHIFT |
451 GEN7_SURFACE_TILING_Y;
453 surf[2] = SET_FIELD(fb->Width - 1, GEN7_SURFACE_WIDTH) |
454 SET_FIELD(fb->Height - 1, GEN7_SURFACE_HEIGHT);
456 gen7_check_surface_setup(surf, true /* is_render_target */);
460 * Sets up a surface state structure to point at the given region.
461 * While it is only used for the front/back buffer currently, it should be
462 * usable for further buffers when doing ARB_draw_buffer support.
465 gen7_update_renderbuffer_surface(struct brw_context *brw,
466 struct gl_renderbuffer *rb,
469 struct intel_context *intel = &brw->intel;
470 struct gl_context *ctx = &intel->ctx;
471 struct intel_renderbuffer *irb = intel_renderbuffer(rb);
472 struct intel_region *region = irb->mt->region;
473 uint32_t tile_x, tile_y;
475 gl_format rb_format = intel_rb_format(irb);
477 uint32_t *surf = brw_state_batch(brw, AUB_TRACE_SURFACE_STATE,
478 8 * 4, 32, &brw->wm.surf_offset[unit]);
479 memset(surf, 0, 8 * 4);
481 /* Render targets can't use IMS layout */
482 assert(irb->mt->msaa_layout != INTEL_MSAA_LAYOUT_IMS);
485 case MESA_FORMAT_SARGB8:
488 * Without GL_EXT_framebuffer_sRGB we shouldn't bind sRGB surfaces to the
489 * blend/update as sRGB.
491 if (ctx->Color.sRGBEnabled)
492 format = brw_format_for_mesa_format(rb_format);
494 format = BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
497 assert(brw_render_target_supported(intel, rb));
498 format = brw->render_target_format[rb_format];
499 if (unlikely(!brw->format_supported_as_render_target[rb_format])) {
500 _mesa_problem(ctx, "%s: renderbuffer format %s unsupported\n",
501 __FUNCTION__, _mesa_get_format_name(rb_format));
506 surf[0] = BRW_SURFACE_2D << BRW_SURFACE_TYPE_SHIFT |
507 format << BRW_SURFACE_FORMAT_SHIFT |
508 (irb->mt->array_spacing_lod0 ? GEN7_SURFACE_ARYSPC_LOD0
509 : GEN7_SURFACE_ARYSPC_FULL) |
510 gen7_surface_tiling_mode(region->tiling);
512 if (irb->mt->align_h == 4)
513 surf[0] |= GEN7_SURFACE_VALIGN_4;
514 if (irb->mt->align_w == 8)
515 surf[0] |= GEN7_SURFACE_HALIGN_8;
518 surf[1] = intel_renderbuffer_tile_offsets(irb, &tile_x, &tile_y) +
519 region->bo->offset; /* reloc */
521 assert(brw->has_surface_tile_offset);
522 /* Note that the low bits of these fields are missing, so
523 * there's the possibility of getting in trouble.
525 assert(tile_x % 4 == 0);
526 assert(tile_y % 2 == 0);
527 surf[5] = SET_FIELD(tile_x / 4, BRW_SURFACE_X_OFFSET) |
528 SET_FIELD(tile_y / 2, BRW_SURFACE_Y_OFFSET);
530 surf[2] = SET_FIELD(rb->Width - 1, GEN7_SURFACE_WIDTH) |
531 SET_FIELD(rb->Height - 1, GEN7_SURFACE_HEIGHT);
532 surf[3] = (region->pitch * region->cpp) - 1;
534 surf[4] = gen7_surface_msaa_bits(irb->mt->num_samples, irb->mt->msaa_layout);
536 if (irb->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS) {
537 gen7_set_surface_mcs_info(brw, surf, brw->wm.surf_offset[unit],
538 irb->mt->mcs_mt, true /* is RT */);
541 if (intel->is_haswell) {
542 surf[7] = SET_FIELD(HSW_SCS_RED, GEN7_SURFACE_SCS_R) |
543 SET_FIELD(HSW_SCS_GREEN, GEN7_SURFACE_SCS_G) |
544 SET_FIELD(HSW_SCS_BLUE, GEN7_SURFACE_SCS_B) |
545 SET_FIELD(HSW_SCS_ALPHA, GEN7_SURFACE_SCS_A);
548 drm_intel_bo_emit_reloc(brw->intel.batch.bo,
549 brw->wm.surf_offset[unit] + 4,
551 surf[1] - region->bo->offset,
552 I915_GEM_DOMAIN_RENDER,
553 I915_GEM_DOMAIN_RENDER);
555 gen7_check_surface_setup(surf, true /* is_render_target */);
559 gen7_init_vtable_surface_functions(struct brw_context *brw)
561 struct intel_context *intel = &brw->intel;
563 intel->vtbl.update_texture_surface = gen7_update_texture_surface;
564 intel->vtbl.update_renderbuffer_surface = gen7_update_renderbuffer_surface;
565 intel->vtbl.update_null_renderbuffer_surface =
566 gen7_update_null_renderbuffer_surface;
567 intel->vtbl.create_constant_surface = gen7_create_constant_surface;