/**************************************************************************
*
* Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
- * Copyright 2009 Intel Corporation.
+ * Copyright 2009, 2012 Intel Corporation.
* All Rights Reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a
#include "swrast/swrast.h"
#include "drivers/common/meta.h"
-#include "intel_context.h"
+#include "intel_batchbuffer.h"
#include "intel_blit.h"
-#include "intel_clear.h"
#include "intel_fbo.h"
+#include "intel_mipmap_tree.h"
#include "intel_regions.h"
+#include "brw_context.h"
+#include "brw_blorp.h"
+
#define FILE_DEBUG_FLAG DEBUG_BLIT
static const char *buffer_names[] = {
}
/**
- * Called by ctx->Driver.Clear.
+ * Returns true if the scissor is a noop (cuts out nothing).
*/
-static void
-intelClear(struct gl_context *ctx, GLbitfield mask)
+static bool
+noop_scissor(struct gl_context *ctx, struct gl_framebuffer *fb)
+{
+ return ctx->Scissor.X <= 0 &&
+ ctx->Scissor.Y <= 0 &&
+ ctx->Scissor.Width >= fb->Width &&
+ ctx->Scissor.Height >= fb->Height;
+}
+
+/**
+ * Implements fast depth clears on gen6+.
+ *
+ * Fast clears basically work by setting a flag in each of the subspans
+ * represented in the HiZ buffer that says "When you need the depth values for
+ * this subspan, it's the hardware's current clear value." Then later rendering
+ * can just use the static clear value instead of referencing memory.
+ *
+ * The tricky part of the implementation is that you have to have the clear
+ * value that was used on the depth buffer in place for all further rendering,
+ * at least until a resolve to the real depth buffer happens.
+ */
+static bool
+brw_fast_clear_depth(struct gl_context *ctx)
{
- struct intel_context *intel = intel_context(ctx);
- const GLuint colorMask = *((GLuint *) & ctx->Color.ColorMask[0]);
- GLbitfield tri_mask = 0;
- GLbitfield blit_mask = 0;
- GLbitfield swrast_mask = 0;
+ struct brw_context *brw = brw_context(ctx);
struct gl_framebuffer *fb = ctx->DrawBuffer;
- struct intel_renderbuffer *irb;
- int i;
+ struct intel_renderbuffer *depth_irb =
+ intel_get_renderbuffer(fb, BUFFER_DEPTH);
+ struct intel_mipmap_tree *mt = depth_irb->mt;
+ struct gl_renderbuffer_attachment *depth_att = &fb->Attachment[BUFFER_DEPTH];
- if (!_mesa_check_conditional_render(ctx))
- return;
+ if (brw->gen < 6)
+ return false;
- if (mask & (BUFFER_BIT_FRONT_LEFT | BUFFER_BIT_FRONT_RIGHT)) {
- intel->front_buffer_dirty = true;
+ if (!intel_renderbuffer_has_hiz(depth_irb))
+ return false;
+
+ /* We only handle full buffer clears -- otherwise you'd have to track whether
+ * a previous clear had happened at a different clear value and resolve it
+ * first.
+ */
+ if (ctx->Scissor.Enabled && !noop_scissor(ctx, fb)) {
+ perf_debug("Failed to fast clear depth due to scissor being enabled. "
+ "Possible 5%% performance win if avoided.\n");
+ return false;
}
- if (0)
- fprintf(stderr, "%s\n", __FUNCTION__);
+ uint32_t depth_clear_value;
+ switch (mt->format) {
+ case MESA_FORMAT_Z32_FLOAT_X24S8:
+ case MESA_FORMAT_S8_Z24:
+ /* From the Sandy Bridge PRM, volume 2 part 1, page 314:
+ *
+ * "[DevSNB+]: Several cases exist where Depth Buffer Clear cannot be
+ * enabled (the legacy method of clearing must be performed):
+ *
+ * - If the depth buffer format is D32_FLOAT_S8X24_UINT or
+ * D24_UNORM_S8_UINT.
+ */
+ return false;
- /* Get SW clears out of the way: Anything without an intel_renderbuffer */
- for (i = 0; i < BUFFER_COUNT; i++) {
- if (!(mask & (1 << i)))
- continue;
+ case MESA_FORMAT_Z32_FLOAT:
+ depth_clear_value = float_as_int(ctx->Depth.Clear);
+ break;
- irb = intel_get_renderbuffer(fb, i);
- if (unlikely(!irb)) {
- swrast_mask |= (1 << i);
- mask &= ~(1 << i);
- }
- }
- if (unlikely(swrast_mask)) {
- debug_mask("swrast", swrast_mask);
- _swrast_Clear(ctx, swrast_mask);
+ case MESA_FORMAT_Z16:
+ /* From the Sandy Bridge PRM, volume 2 part 1, page 314:
+ *
+ * "[DevSNB+]: Several cases exist where Depth Buffer Clear cannot be
+ * enabled (the legacy method of clearing must be performed):
+ *
+ * - DevSNB{W/A}]: When depth buffer format is D16_UNORM and the
+ * width of the map (LOD0) is not multiple of 16, fast clear
+ * optimization must be disabled.
+ */
+ if (brw->gen == 6 && (mt->level[depth_irb->mt_level].width % 16) != 0)
+ return false;
+ /* FALLTHROUGH */
+
+ default:
+ depth_clear_value = fb->_DepthMax * ctx->Depth.Clear;
+ break;
}
- /* HW color buffers (front, back, aux, generic FBO, etc) */
- if (intel->gen < 6 && colorMask == ~0) {
- /* clear all R,G,B,A */
- blit_mask |= (mask & BUFFER_BITS_COLOR);
+ /* If we're clearing to a new clear value, then we need to resolve any clear
+ * flags out of the HiZ buffer into the real depth buffer.
+ */
+ if (mt->depth_clear_value != depth_clear_value) {
+ intel_miptree_all_slices_resolve_depth(brw, mt);
+ mt->depth_clear_value = depth_clear_value;
}
- else {
- /* glColorMask in effect */
- tri_mask |= (mask & BUFFER_BITS_COLOR);
+
+ /* From the Sandy Bridge PRM, volume 2 part 1, page 313:
+ *
+ * "If other rendering operations have preceded this clear, a
+ * PIPE_CONTROL with write cache flush enabled and Z-inhibit disabled
+ * must be issued before the rectangle primitive used for the depth
+ * buffer clear operation.
+ */
+ intel_batchbuffer_emit_mi_flush(brw);
+
+ intel_hiz_exec(brw, mt, depth_irb->mt_level, depth_irb->mt_layer,
+ GEN6_HIZ_OP_DEPTH_CLEAR);
+
+ if (brw->gen == 6) {
+ /* From the Sandy Bridge PRM, volume 2 part 1, page 314:
+ *
+ * "DevSNB, DevSNB-B{W/A}]: Depth buffer clear pass must be followed
+ * by a PIPE_CONTROL command with DEPTH_STALL bit set and Then
+ * followed by Depth FLUSH'
+ */
+ intel_batchbuffer_emit_mi_flush(brw);
}
- /* Make sure we have up to date buffers before we start looking at
- * the tiling bits to determine how to clear. */
- intel_prepare_render(intel);
+ /* Now, the HiZ buffer contains data that needs to be resolved to the depth
+ * buffer.
+ */
+ intel_renderbuffer_att_set_needs_depth_resolve(depth_att);
- /* HW stencil */
- if (mask & BUFFER_BIT_STENCIL) {
- const struct intel_region *stencilRegion
- = intel_get_rb_region(fb, BUFFER_STENCIL);
- if (stencilRegion) {
- tri_mask |= BUFFER_BIT_STENCIL;
- }
+ return true;
+}
+
+/**
+ * Called by ctx->Driver.Clear.
+ */
+static void
+brw_clear(struct gl_context *ctx, GLbitfield mask)
+{
+ struct brw_context *brw = brw_context(ctx);
+ struct gl_framebuffer *fb = ctx->DrawBuffer;
+ bool partial_clear = ctx->Scissor.Enabled && !noop_scissor(ctx, fb);
+
+ if (!_mesa_check_conditional_render(ctx))
+ return;
+
+ if (mask & (BUFFER_BIT_FRONT_LEFT | BUFFER_BIT_FRONT_RIGHT)) {
+ brw->front_buffer_dirty = true;
}
- /* HW depth */
+ intel_prepare_render(brw);
+ brw_workaround_depthstencil_alignment(brw, partial_clear ? 0 : mask);
+
if (mask & BUFFER_BIT_DEPTH) {
- tri_mask |= BUFFER_BIT_DEPTH;
+ if (brw_fast_clear_depth(ctx)) {
+ DBG("fast clear: depth\n");
+ mask &= ~BUFFER_BIT_DEPTH;
+ }
}
- /* If we're doing a tri pass for depth/stencil, include a likely color
- * buffer with it.
- */
- if (mask & (BUFFER_BIT_DEPTH | BUFFER_BIT_STENCIL)) {
- int color_bit = ffs(mask & BUFFER_BITS_COLOR);
- if (color_bit != 0) {
- tri_mask |= blit_mask & (1 << (color_bit - 1));
- blit_mask &= ~(1 << (color_bit - 1));
+ /* BLORP is currently only supported on Gen6+. */
+ if (brw->gen >= 6) {
+ if (mask & BUFFER_BITS_COLOR) {
+ if (brw_blorp_clear_color(brw, fb, partial_clear)) {
+ debug_mask("blorp color", mask & BUFFER_BITS_COLOR);
+ mask &= ~BUFFER_BITS_COLOR;
+ }
}
}
- /* Anything left, just use tris */
- tri_mask |= mask & ~blit_mask;
-
- if (blit_mask) {
- debug_mask("blit", blit_mask);
- tri_mask |= intelClearWithBlit(ctx, blit_mask);
- }
+ GLbitfield tri_mask = mask & (BUFFER_BITS_COLOR |
+ BUFFER_BIT_STENCIL |
+ BUFFER_BIT_DEPTH);
if (tri_mask) {
debug_mask("tri", tri_mask);
- _mesa_meta_glsl_Clear(&intel->ctx, tri_mask);
+ mask &= ~tri_mask;
+
+ if (ctx->API == API_OPENGLES) {
+ _mesa_meta_Clear(&brw->ctx, tri_mask);
+ } else {
+ _mesa_meta_glsl_Clear(&brw->ctx, tri_mask);
+ }
+ }
+
+ /* Any strange buffers get passed off to swrast */
+ if (mask) {
+ debug_mask("swrast", mask);
+ _swrast_Clear(ctx, mask);
}
}
void
intelInitClearFuncs(struct dd_function_table *functions)
{
- functions->Clear = intelClear;
+ functions->Clear = brw_clear;
}