* IN THE SOFTWARE.
*/
+#include "main/mtypes.h"
#include "intel_batchbuffer.h"
#include "intel_mipmap_tree.h"
#include "intel_regions.h"
#include "brw_state.h"
#include "brw_defines.h"
-static void emit_depthbuffer(struct brw_context *brw)
+void
+gen7_emit_depth_stencil_hiz(struct brw_context *brw,
+ struct intel_mipmap_tree *depth_mt,
+ uint32_t depth_offset, uint32_t depthbuffer_format,
+ uint32_t depth_surface_type,
+ struct intel_mipmap_tree *stencil_mt,
+ bool hiz, bool separate_stencil,
+ uint32_t width, uint32_t height,
+ uint32_t tile_x, uint32_t tile_y)
{
- struct intel_context *intel = &brw->intel;
- struct gl_context *ctx = &intel->ctx;
+ struct gl_context *ctx = &brw->ctx;
+ const uint8_t mocs = GEN7_MOCS_L3;
struct gl_framebuffer *fb = ctx->DrawBuffer;
-
- /* _NEW_BUFFERS */
- struct intel_renderbuffer *drb = intel_get_renderbuffer(fb, BUFFER_DEPTH);
- struct intel_renderbuffer *srb = intel_get_renderbuffer(fb, BUFFER_STENCIL);
- struct intel_mipmap_tree *depth_mt = NULL,
- *stencil_mt = NULL,
- *hiz_mt = NULL;
-
- /* Amount by which drawing should be offset in order to draw to the
- * appropriate miplevel/zoffset/cubeface. We will extract these values
- * from depth_irb or stencil_irb once we determine which is present.
- */
- uint32_t draw_x = 0, draw_y = 0;
-
- /* Masks used to determine how much of the draw_x and draw_y offsets should
- * be performed using the fine adjustment of "depth coordinate offset X/Y"
- * (dw5 of 3DSTATE_DEPTH_BUFFER). Any remaining coarse adjustment will be
- * performed by changing the base addresses of the buffers.
- *
- * Since the HiZ, depth, and stencil buffers all use the same "depth
- * coordinate offset X/Y" values, we need to make sure that the coarse
- * adjustment will be possible to apply to all three buffers. Since coarse
- * adjustment can only be applied in multiples of the tile size, we will OR
- * together the tile masks of all the buffers to determine which offsets to
- * perform as fine adjustments.
- */
- uint32_t tile_mask_x = 0, tile_mask_y = 0;
-
- if (drb)
- depth_mt = drb->mt;
-
- if (depth_mt) {
- hiz_mt = depth_mt->hiz_mt;
-
- intel_region_get_tile_masks(depth_mt->region,
- &tile_mask_x, &tile_mask_y, false);
-
- if (hiz_mt) {
- uint32_t hiz_tile_mask_x, hiz_tile_mask_y;
- intel_region_get_tile_masks(hiz_mt->region,
- &hiz_tile_mask_x, &hiz_tile_mask_y,
- false);
-
- /* Each HiZ row represents 2 rows of pixels */
- hiz_tile_mask_y = hiz_tile_mask_y << 1 | 1;
-
- tile_mask_x |= hiz_tile_mask_x;
- tile_mask_y |= hiz_tile_mask_y;
- }
+ uint32_t surftype;
+ unsigned int depth = 1;
+ unsigned int min_array_element;
+ GLenum gl_target = GL_TEXTURE_2D;
+ unsigned int lod;
+ const struct intel_mipmap_tree *mt = depth_mt ? depth_mt : stencil_mt;
+ const struct intel_renderbuffer *irb = NULL;
+ const struct gl_renderbuffer *rb = NULL;
+
+ intel_emit_depth_stall_flushes(brw);
+
+ irb = intel_get_renderbuffer(fb, BUFFER_DEPTH);
+ if (!irb)
+ irb = intel_get_renderbuffer(fb, BUFFER_STENCIL);
+ rb = (struct gl_renderbuffer*) irb;
+
+ if (rb) {
+ depth = MAX2(rb->Depth, 1);
+ if (rb->TexImage)
+ gl_target = rb->TexImage->TexObject->Target;
}
- if (srb) {
- stencil_mt = srb->mt;
- if (stencil_mt->stencil_mt)
- stencil_mt = stencil_mt->stencil_mt;
-
- assert(stencil_mt->format == MESA_FORMAT_S8);
-
- /* Stencil buffer uses 64x64 tiles. */
- tile_mask_x |= 63;
- tile_mask_y |= 63;
+ switch (gl_target) {
+ case GL_TEXTURE_CUBE_MAP_ARRAY:
+ case GL_TEXTURE_CUBE_MAP:
+ /* The PRM claims that we should use BRW_SURFACE_CUBE for this
+ * situation, but experiments show that gl_Layer doesn't work when we do
+ * this. So we use BRW_SURFACE_2D, since for rendering purposes this is
+ * equivalent.
+ */
+ surftype = BRW_SURFACE_2D;
+ depth *= 6;
+ break;
+ default:
+ surftype = translate_tex_target(gl_target);
+ break;
}
- /* Gen7 doesn't support packed depth/stencil */
- assert(stencil_mt == NULL || depth_mt != stencil_mt);
- assert(!depth_mt || !_mesa_is_format_packed_depth_stencil(depth_mt->format));
-
- intel_emit_depth_stall_flushes(intel);
-
- if (depth_mt == NULL) {
- uint32_t dw1 = BRW_DEPTHFORMAT_D32_FLOAT << 18;
- uint32_t dw3 = 0;
- uint32_t tile_x = 0, tile_y = 0;
+ if (fb->Layered || !irb) {
+ min_array_element = 0;
+ } else if (irb->mt->num_samples > 1) {
+ /* Convert physical layer to logical layer. */
+ min_array_element = irb->mt_layer / irb->mt->num_samples;
+ } else {
+ min_array_element = irb->mt_layer;
+ }
- if (stencil_mt == NULL) {
- dw1 |= (BRW_SURFACE_NULL << 29);
- } else {
- /* _NEW_STENCIL: enable stencil buffer writes */
- dw1 |= ((ctx->Stencil.WriteMask != 0) << 27);
+ lod = irb ? irb->mt_level - irb->mt->first_level : 0;
- draw_x = srb->draw_x;
- draw_y = srb->draw_y;
- tile_x = draw_x & tile_mask_x;
- tile_y = draw_y & tile_mask_y;
+ if (mt) {
+ width = mt->physical_width0;
+ height = mt->physical_height0;
+ }
- /* According to the Sandy Bridge PRM, volume 2 part 1, pp326-327
- * (3DSTATE_DEPTH_BUFFER dw5), in the documentation for "Depth
- * Coordinate Offset X/Y":
- *
- * "The 3 LSBs of both offsets must be zero to ensure correct
- * alignment"
- *
- * We have no guarantee that tile_x and tile_y are correctly aligned,
- * since they are determined by the mipmap layout, which is only
- * aligned to multiples of 4.
- *
- * So, to avoid hanging the GPU, just smash the low order 3 bits of
- * tile_x and tile_y to 0. This is a temporary workaround until we
- * come up with a better solution.
- */
- tile_x &= ~7;
- tile_y &= ~7;
+ /* _NEW_DEPTH, _NEW_STENCIL, _NEW_BUFFERS */
+ BEGIN_BATCH(7);
+ /* 3DSTATE_DEPTH_BUFFER dw0 */
+ OUT_BATCH(GEN7_3DSTATE_DEPTH_BUFFER << 16 | (7 - 2));
- /* 3DSTATE_STENCIL_BUFFER inherits surface type and dimensions. */
- dw1 |= (BRW_SURFACE_2D << 29);
- dw3 = ((srb->Base.Base.Width + tile_x - 1) << 4) |
- ((srb->Base.Base.Height + tile_y - 1) << 18);
- }
+ /* 3DSTATE_DEPTH_BUFFER dw1 */
+ OUT_BATCH((depth_mt ? depth_mt->region->pitch - 1 : 0) |
+ (depthbuffer_format << 18) |
+ ((hiz ? 1 : 0) << 22) |
+ ((stencil_mt != NULL && ctx->Stencil._WriteEnabled) << 27) |
+ ((ctx->Depth.Mask != 0) << 28) |
+ (surftype << 29));
- BEGIN_BATCH(7);
- OUT_BATCH(GEN7_3DSTATE_DEPTH_BUFFER << 16 | (7 - 2));
- OUT_BATCH(dw1);
- OUT_BATCH(0);
- OUT_BATCH(dw3);
- OUT_BATCH(0);
- OUT_BATCH(tile_x | (tile_y << 16));
- OUT_BATCH(0);
- ADVANCE_BATCH();
+ /* 3DSTATE_DEPTH_BUFFER dw2 */
+ if (depth_mt) {
+ OUT_RELOC(depth_mt->region->bo,
+ I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
+ 0);
} else {
- struct intel_region *region = depth_mt->region;
- uint32_t tile_x, tile_y, offset;
+ OUT_BATCH(0);
+ }
- draw_x = drb->draw_x;
- draw_y = drb->draw_y;
- tile_x = draw_x & tile_mask_x;
- tile_y = draw_y & tile_mask_y;
+ /* 3DSTATE_DEPTH_BUFFER dw3 */
+ OUT_BATCH(((width - 1) << 4) |
+ ((height - 1) << 18) |
+ lod);
- /* According to the Sandy Bridge PRM, volume 2 part 1, pp326-327
- * (3DSTATE_DEPTH_BUFFER dw5), in the documentation for "Depth
- * Coordinate Offset X/Y":
- *
- * "The 3 LSBs of both offsets must be zero to ensure correct
- * alignment"
- *
- * We have no guarantee that tile_x and tile_y are correctly aligned,
- * since they are determined by the mipmap layout, which is only aligned
- * to multiples of 4.
- *
- * So, to avoid hanging the GPU, just smash the low order 3 bits of
- * tile_x and tile_y to 0. This is a temporary workaround until we come
- * up with a better solution.
- */
- tile_x &= ~7;
- tile_y &= ~7;
+ /* 3DSTATE_DEPTH_BUFFER dw4 */
+ OUT_BATCH(((depth - 1) << 21) |
+ (min_array_element << 10) |
+ mocs);
- offset = intel_region_get_aligned_offset(region,
- draw_x & ~tile_mask_x,
- draw_y & ~tile_mask_y);
+ /* 3DSTATE_DEPTH_BUFFER dw5 */
+ OUT_BATCH(0);
- assert(region->tiling == I915_TILING_Y);
-
- /* _NEW_DEPTH, _NEW_STENCIL */
- BEGIN_BATCH(7);
- OUT_BATCH(GEN7_3DSTATE_DEPTH_BUFFER << 16 | (7 - 2));
- OUT_BATCH(((region->pitch * region->cpp) - 1) |
- (brw_depthbuffer_format(brw) << 18) |
- ((hiz_mt ? 1 : 0) << 22) | /* hiz enable */
- ((stencil_mt != NULL && ctx->Stencil.WriteMask != 0) << 27) |
- ((ctx->Depth.Mask != 0) << 28) |
- (BRW_SURFACE_2D << 29));
- OUT_RELOC(region->bo,
- I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
- offset);
- OUT_BATCH((((drb->Base.Base.Width + tile_x) - 1) << 4) |
- (((drb->Base.Base.Height + tile_y) - 1) << 18));
- OUT_BATCH(0);
- OUT_BATCH(tile_x | (tile_y << 16));
- OUT_BATCH(0);
- ADVANCE_BATCH();
- }
+ /* 3DSTATE_DEPTH_BUFFER dw6 */
+ OUT_BATCH((depth - 1) << 21);
+ ADVANCE_BATCH();
- if (hiz_mt == NULL) {
+ if (!hiz) {
BEGIN_BATCH(3);
OUT_BATCH(GEN7_3DSTATE_HIER_DEPTH_BUFFER << 16 | (3 - 2));
OUT_BATCH(0);
OUT_BATCH(0);
ADVANCE_BATCH();
} else {
- uint32_t hiz_offset =
- intel_region_get_aligned_offset(hiz_mt->region,
- draw_x & ~tile_mask_x,
- (draw_y & ~tile_mask_y) / 2);
+ struct intel_mipmap_tree *hiz_mt = depth_mt->hiz_mt;
BEGIN_BATCH(3);
OUT_BATCH(GEN7_3DSTATE_HIER_DEPTH_BUFFER << 16 | (3 - 2));
- OUT_BATCH(hiz_mt->region->pitch * hiz_mt->region->cpp - 1);
+ OUT_BATCH((mocs << 25) |
+ (hiz_mt->region->pitch - 1));
OUT_RELOC(hiz_mt->region->bo,
I915_GEM_DOMAIN_RENDER,
I915_GEM_DOMAIN_RENDER,
- hiz_offset);
+ 0);
ADVANCE_BATCH();
}
OUT_BATCH(0);
ADVANCE_BATCH();
} else {
- const int enabled = intel->is_haswell ? HSW_STENCIL_ENABLED : 0;
-
- /* Note: We can't compute the stencil offset using
- * intel_region_get_aligned_offset(), because the stencil region claims
- * that the region is untiled; in fact it's W tiled.
- */
- uint32_t stencil_offset =
- (draw_y & ~tile_mask_y) * stencil_mt->region->pitch +
- (draw_x & ~tile_mask_x) * 64;
+ const int enabled = brw->is_haswell ? HSW_STENCIL_ENABLED : 0;
BEGIN_BATCH(3);
OUT_BATCH(GEN7_3DSTATE_STENCIL_BUFFER << 16 | (3 - 2));
- /* The stencil buffer has quirky pitch requirements. From the Graphics
- * BSpec: vol2a.11 3D Pipeline Windower > Early Depth/Stencil Processing
- * > Depth/Stencil Buffer State > 3DSTATE_STENCIL_BUFFER [DevIVB+],
- * field "Surface Pitch":
+ /* The stencil buffer has quirky pitch requirements. From the
+ * Sandybridge PRM, Volume 2 Part 1, page 329 (3DSTATE_STENCIL_BUFFER
+ * dword 1 bits 16:0 - Surface Pitch):
*
* The pitch must be set to 2x the value computed based on width, as
* the stencil buffer is stored with two rows interleaved.
*
- * (Note that it is not 100% clear whether this intended to apply to
- * Gen7; the BSpec flags this comment as "DevILK,DevSNB" (which would
- * imply that it doesn't), however the comment appears on a "DevIVB+"
- * page (which would imply that it does). Experiments with the hardware
- * indicate that it does.
+ * While the Ivybridge PRM lacks this comment, the BSpec contains the
+ * same text, and experiments indicate that this is necessary.
*/
OUT_BATCH(enabled |
- (2 * stencil_mt->region->pitch * stencil_mt->region->cpp - 1));
+ mocs << 25 |
+ (2 * stencil_mt->region->pitch - 1));
OUT_RELOC(stencil_mt->region->bo,
I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
- stencil_offset);
+ 0);
ADVANCE_BATCH();
}
.brw = BRW_NEW_BATCH,
.cache = 0,
},
- .emit = emit_depthbuffer,
+ .emit = brw_emit_depthbuffer,
};