1 /************************************************************************** 2 * 3 * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas. 4 * Copyright 2009, 2012 Intel Corporation. 5 * All Rights Reserved. 6 * 7 * Permission is hereby granted, free of charge, to any person obtaining a 8 * copy of this software and associated documentation files (the 9 * "Software"), to deal in the Software without restriction, including 10 * without limitation the rights to use, copy, modify, merge, publish, 11 * distribute, sub license, and/or sell copies of the Software, and to 12 * permit persons to whom the Software is furnished to do so, subject to 13 * the following conditions: 14 * 15 * The above copyright notice and this permission notice (including the 16 * next paragraph) shall be included in all copies or substantial portions 17 * of the Software. 18 * 19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. 22 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR 23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, 24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE 25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 26 * 27 **************************************************************************/ 28 29 #include "main/glheader.h" 30 #include "main/mtypes.h" 31 #include "main/condrender.h" 32 #include "swrast/swrast.h" 33 #include "drivers/common/meta.h" 34 35 #include "intel_batchbuffer.h" 36 #include "intel_context.h" 37 #include "intel_blit.h" 38 #include "intel_clear.h" 39 #include "intel_fbo.h" 40 #include "intel_mipmap_tree.h" 41 #include "intel_regions.h" 42 43 #define FILE_DEBUG_FLAG DEBUG_BLIT 44 45 static const char *buffer_names[] = { 46 [BUFFER_FRONT_LEFT] = "front", 47 [BUFFER_BACK_LEFT] = "back", 48 [BUFFER_FRONT_RIGHT] = "front right", 49 [BUFFER_BACK_RIGHT] = "back right", 50 [BUFFER_DEPTH] = "depth", 51 [BUFFER_STENCIL] = "stencil", 52 [BUFFER_ACCUM] = "accum", 53 [BUFFER_AUX0] = "aux0", 54 [BUFFER_COLOR0] = "color0", 55 [BUFFER_COLOR1] = "color1", 56 [BUFFER_COLOR2] = "color2", 57 [BUFFER_COLOR3] = "color3", 58 [BUFFER_COLOR4] = "color4", 59 [BUFFER_COLOR5] = "color5", 60 [BUFFER_COLOR6] = "color6", 61 [BUFFER_COLOR7] = "color7", 62 }; 63 64 static void 65 debug_mask(const char *name, GLbitfield mask) 66 { 67 GLuint i; 68 69 if (unlikely(INTEL_DEBUG & DEBUG_BLIT)) { 70 DBG("%s clear:", name); 71 for (i = 0; i < BUFFER_COUNT; i++) { 72 if (mask & (1 << i)) 73 DBG(" %s", buffer_names[i]); 74 } 75 DBG("\n"); 76 } 77 } 78 79 /** 80 * Implements fast depth clears on gen6+. 81 * 82 * Fast clears basically work by setting a flag in each of the subspans 83 * represented in the HiZ buffer that says "When you need the depth values for 84 * this subspan, it's the hardware's current clear value." Then later rendering 85 * can just use the static clear value instead of referencing memory. 86 * 87 * The tricky part of the implementation is that you have to have the clear 88 * value that was used on the depth buffer in place for all further rendering, 89 * at least until a resolve to the real depth buffer happens. 90 */ 91 static bool 92 brw_fast_clear_depth(struct gl_context *ctx) 93 { 94 struct intel_context *intel = intel_context(ctx); 95 struct gl_framebuffer *fb = ctx->DrawBuffer; 96 struct intel_renderbuffer *depth_irb = 97 intel_get_renderbuffer(fb, BUFFER_DEPTH); 98 struct intel_mipmap_tree *mt = depth_irb->mt; 99 100 if (intel->gen < 6) 101 return false; 102 103 if (!mt->hiz_mt) 104 return false; 105 106 /* We only handle full buffer clears -- otherwise you'd have to track whether 107 * a previous clear had happened at a different clear value and resolve it 108 * first. 109 */ 110 if (ctx->Scissor.Enabled) { 111 perf_debug("Failed to fast clear depth due to scissor being enabled. " 112 "Possible 5%% performance win if avoided.\n"); 113 return false; 114 } 115 116 /* The rendered area has to be 8x4 samples, not resolved pixels, so we look 117 * at the miptree slice dimensions instead of renderbuffer size. 118 */ 119 if (mt->level[depth_irb->mt_level].width % 8 != 0 || 120 mt->level[depth_irb->mt_level].height % 4 != 0) { 121 perf_debug("Failed to fast clear depth due to width/height %d,%d not " 122 "being aligned to 8,4. Possible 5%% performance win if " 123 "avoided\n", 124 mt->level[depth_irb->mt_level].width, 125 mt->level[depth_irb->mt_level].height); 126 return false; 127 } 128 129 uint32_t depth_clear_value; 130 switch (mt->format) { 131 case MESA_FORMAT_Z32_FLOAT_X24S8: 132 case MESA_FORMAT_S8_Z24: 133 /* From the Sandy Bridge PRM, volume 2 part 1, page 314: 134 * 135 * "[DevSNB+]: Several cases exist where Depth Buffer Clear cannot be 136 * enabled (the legacy method of clearing must be performed): 137 * 138 * - If the depth buffer format is D32_FLOAT_S8X24_UINT or 139 * D24_UNORM_S8_UINT. 140 */ 141 return false; 142 143 case MESA_FORMAT_Z32_FLOAT: 144 depth_clear_value = float_as_int(ctx->Depth.Clear); 145 break; 146 147 case MESA_FORMAT_Z16: 148 /* From the Sandy Bridge PRM, volume 2 part 1, page 314: 149 * 150 * "[DevSNB+]: Several cases exist where Depth Buffer Clear cannot be 151 * enabled (the legacy method of clearing must be performed): 152 * 153 * - DevSNB{W/A}]: When depth buffer format is D16_UNORM and the 154 * width of the map (LOD0) is not multiple of 16, fast clear 155 * optimization must be disabled. 156 */ 157 if (intel->gen == 6 && (mt->level[depth_irb->mt_level].width % 16) != 0) 158 return false; 159 /* FALLTHROUGH */ 160 161 default: 162 depth_clear_value = fb->_DepthMax * ctx->Depth.Clear; 163 break; 164 } 165 166 /* If we're clearing to a new clear value, then we need to resolve any clear 167 * flags out of the HiZ buffer into the real depth buffer. 168 */ 169 if (mt->depth_clear_value != depth_clear_value) { 170 intel_miptree_all_slices_resolve_depth(intel, mt); 171 mt->depth_clear_value = depth_clear_value; 172 } 173 174 /* From the Sandy Bridge PRM, volume 2 part 1, page 313: 175 * 176 * "If other rendering operations have preceded this clear, a 177 * PIPE_CONTROL with write cache flush enabled and Z-inhibit disabled 178 * must be issued before the rectangle primitive used for the depth 179 * buffer clear operation. 180 */ 181 intel_batchbuffer_emit_mi_flush(intel); 182 183 intel_hiz_exec(intel, mt, depth_irb->mt_level, depth_irb->mt_layer, 184 GEN6_HIZ_OP_DEPTH_CLEAR); 185 186 if (intel->gen == 6) { 187 /* From the Sandy Bridge PRM, volume 2 part 1, page 314: 188 * 189 * "DevSNB, DevSNB-B{W/A}]: Depth buffer clear pass must be followed 190 * by a PIPE_CONTROL command with DEPTH_STALL bit set and Then 191 * followed by Depth FLUSH' 192 */ 193 intel_batchbuffer_emit_mi_flush(intel); 194 } 195 196 /* Now, the HiZ buffer contains data that needs to be resolved to the depth 197 * buffer. 198 */ 199 intel_renderbuffer_set_needs_depth_resolve(depth_irb); 200 201 return true; 202 } 203 204 /** 205 * Called by ctx->Driver.Clear. 206 */ 207 static void 208 brw_clear(struct gl_context *ctx, GLbitfield mask) 209 { 210 struct intel_context *intel = intel_context(ctx); 211 212 if (!_mesa_check_conditional_render(ctx)) 213 return; 214 215 if (mask & (BUFFER_BIT_FRONT_LEFT | BUFFER_BIT_FRONT_RIGHT)) { 216 intel->front_buffer_dirty = true; 217 } 218 219 intel_prepare_render(intel); 220 221 if (mask & BUFFER_BIT_DEPTH) { 222 if (brw_fast_clear_depth(ctx)) { 223 DBG("fast clear: depth\n"); 224 mask &= ~BUFFER_BIT_DEPTH; 225 } 226 } 227 228 GLbitfield tri_mask = mask & (BUFFER_BITS_COLOR | 229 BUFFER_BIT_STENCIL | 230 BUFFER_BIT_DEPTH); 231 232 if (tri_mask) { 233 debug_mask("tri", tri_mask); 234 mask &= ~tri_mask; 235 236 if (ctx->API == API_OPENGLES) { 237 _mesa_meta_Clear(&intel->ctx, tri_mask); 238 } else { 239 _mesa_meta_glsl_Clear(&intel->ctx, tri_mask); 240 } 241 } 242 243 /* Any strange buffers get passed off to swrast */ 244 if (mask) { 245 debug_mask("swrast", mask); 246 _swrast_Clear(ctx, mask); 247 } 248 } 249 250 251 void 252 intelInitClearFuncs(struct dd_function_table *functions) 253 { 254 functions->Clear = brw_clear; 255 } 256
