1 /* 2 * Copyright 2015 Intel Corporation 3 * 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: 10 * 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 13 * Software. 14 * 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 21 * IN THE SOFTWARE. 22 */ 23 24 #include <assert.h> 25 #include <stdbool.h> 26 #include <string.h> 27 #include <unistd.h> 28 #include <fcntl.h> 29 30 #include "anv_private.h" 31 32 #include "genxml/gen_macros.h" 33 #include "genxml/genX_pack.h" 34 35 #if GEN_GEN == 8 36 void 37 gen8_cmd_buffer_emit_viewport(struct anv_cmd_buffer *cmd_buffer) 38 { 39 uint32_t count = cmd_buffer->state.dynamic.viewport.count; 40 const VkViewport *viewports = cmd_buffer->state.dynamic.viewport.viewports; 41 struct anv_state sf_clip_state = 42 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, count * 64, 64); 43 44 for (uint32_t i = 0; i < count; i++) { 45 const VkViewport *vp = &viewports[i]; 46 47 /* The gen7 state struct has just the matrix and guardband fields, the 48 * gen8 struct adds the min/max viewport fields. */ 49 struct GENX(SF_CLIP_VIEWPORT) sf_clip_viewport = { 50 .ViewportMatrixElementm00 = vp->width / 2, 51 .ViewportMatrixElementm11 = vp->height / 2, 52 .ViewportMatrixElementm22 = 1.0, 53 .ViewportMatrixElementm30 = vp->x + vp->width / 2, 54 .ViewportMatrixElementm31 = vp->y + vp->height / 2, 55 .ViewportMatrixElementm32 = 0.0, 56 .XMinClipGuardband = -1.0f, 57 .XMaxClipGuardband = 1.0f, 58 .YMinClipGuardband = -1.0f, 59 .YMaxClipGuardband = 1.0f, 60 .XMinViewPort = vp->x, 61 .XMaxViewPort = vp->x + vp->width - 1, 62 .YMinViewPort = vp->y, 63 .YMaxViewPort = vp->y + vp->height - 1, 64 }; 65 66 GENX(SF_CLIP_VIEWPORT_pack)(NULL, sf_clip_state.map + i * 64, 67 &sf_clip_viewport); 68 } 69 70 if (!cmd_buffer->device->info.has_llc) 71 anv_state_clflush(sf_clip_state); 72 73 anv_batch_emit(&cmd_buffer->batch, 74 GENX(3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP), clip) { 75 clip.SFClipViewportPointer = sf_clip_state.offset; 76 } 77 } 78 79 void 80 gen8_cmd_buffer_emit_depth_viewport(struct anv_cmd_buffer *cmd_buffer, 81 bool depth_clamp_enable) 82 { 83 uint32_t count = cmd_buffer->state.dynamic.viewport.count; 84 const VkViewport *viewports = cmd_buffer->state.dynamic.viewport.viewports; 85 struct anv_state cc_state = 86 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, count * 8, 32); 87 88 for (uint32_t i = 0; i < count; i++) { 89 const VkViewport *vp = &viewports[i]; 90 91 struct GENX(CC_VIEWPORT) cc_viewport = { 92 .MinimumDepth = depth_clamp_enable ? vp->minDepth : 0.0f, 93 .MaximumDepth = depth_clamp_enable ? vp->maxDepth : 1.0f, 94 }; 95 96 GENX(CC_VIEWPORT_pack)(NULL, cc_state.map + i * 8, &cc_viewport); 97 } 98 99 if (!cmd_buffer->device->info.has_llc) 100 anv_state_clflush(cc_state); 101 102 anv_batch_emit(&cmd_buffer->batch, 103 GENX(3DSTATE_VIEWPORT_STATE_POINTERS_CC), cc) { 104 cc.CCViewportPointer = cc_state.offset; 105 } 106 } 107 #endif 108 109 static void 110 __emit_genx_sf_state(struct anv_cmd_buffer *cmd_buffer) 111 { 112 uint32_t sf_dw[GENX(3DSTATE_SF_length)]; 113 struct GENX(3DSTATE_SF) sf = { 114 GENX(3DSTATE_SF_header), 115 .LineWidth = cmd_buffer->state.dynamic.line_width, 116 }; 117 GENX(3DSTATE_SF_pack)(NULL, sf_dw, &sf); 118 /* FIXME: gen9.fs */ 119 anv_batch_emit_merge(&cmd_buffer->batch, sf_dw, 120 cmd_buffer->state.pipeline->gen8.sf); 121 } 122 123 void 124 gen9_emit_sf_state(struct anv_cmd_buffer *cmd_buffer); 125 126 #if GEN_GEN == 9 127 128 void 129 gen9_emit_sf_state(struct anv_cmd_buffer *cmd_buffer) 130 { 131 __emit_genx_sf_state(cmd_buffer); 132 } 133 134 #endif 135 136 #if GEN_GEN == 8 137 138 static void 139 __emit_sf_state(struct anv_cmd_buffer *cmd_buffer) 140 { 141 if (cmd_buffer->device->info.is_cherryview) 142 gen9_emit_sf_state(cmd_buffer); 143 else 144 __emit_genx_sf_state(cmd_buffer); 145 } 146 147 #else 148 149 static void 150 __emit_sf_state(struct anv_cmd_buffer *cmd_buffer) 151 { 152 __emit_genx_sf_state(cmd_buffer); 153 } 154 155 #endif 156 157 void 158 genX(cmd_buffer_flush_dynamic_state)(struct anv_cmd_buffer *cmd_buffer) 159 { 160 struct anv_pipeline *pipeline = cmd_buffer->state.pipeline; 161 162 if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_PIPELINE | 163 ANV_CMD_DIRTY_DYNAMIC_LINE_WIDTH)) { 164 __emit_sf_state(cmd_buffer); 165 } 166 167 if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_PIPELINE | 168 ANV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS)){ 169 uint32_t raster_dw[GENX(3DSTATE_RASTER_length)]; 170 struct GENX(3DSTATE_RASTER) raster = { 171 GENX(3DSTATE_RASTER_header), 172 .GlobalDepthOffsetConstant = cmd_buffer->state.dynamic.depth_bias.bias, 173 .GlobalDepthOffsetScale = cmd_buffer->state.dynamic.depth_bias.slope, 174 .GlobalDepthOffsetClamp = cmd_buffer->state.dynamic.depth_bias.clamp 175 }; 176 GENX(3DSTATE_RASTER_pack)(NULL, raster_dw, &raster); 177 anv_batch_emit_merge(&cmd_buffer->batch, raster_dw, 178 pipeline->gen8.raster); 179 } 180 181 /* Stencil reference values moved from COLOR_CALC_STATE in gen8 to 182 * 3DSTATE_WM_DEPTH_STENCIL in gen9. That means the dirty bits gets split 183 * across different state packets for gen8 and gen9. We handle that by 184 * using a big old #if switch here. 185 */ 186 #if GEN_GEN == 8 187 if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS | 188 ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE)) { 189 struct anv_dynamic_state *d = &cmd_buffer->state.dynamic; 190 struct anv_state cc_state = 191 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, 192 GENX(COLOR_CALC_STATE_length) * 4, 193 64); 194 struct GENX(COLOR_CALC_STATE) cc = { 195 .BlendConstantColorRed = cmd_buffer->state.dynamic.blend_constants[0], 196 .BlendConstantColorGreen = cmd_buffer->state.dynamic.blend_constants[1], 197 .BlendConstantColorBlue = cmd_buffer->state.dynamic.blend_constants[2], 198 .BlendConstantColorAlpha = cmd_buffer->state.dynamic.blend_constants[3], 199 .StencilReferenceValue = d->stencil_reference.front & 0xff, 200 .BackFaceStencilReferenceValue = d->stencil_reference.back & 0xff, 201 }; 202 GENX(COLOR_CALC_STATE_pack)(NULL, cc_state.map, &cc); 203 204 if (!cmd_buffer->device->info.has_llc) 205 anv_state_clflush(cc_state); 206 207 anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_CC_STATE_POINTERS), ccp) { 208 ccp.ColorCalcStatePointer = cc_state.offset; 209 ccp.ColorCalcStatePointerValid = true; 210 } 211 } 212 213 if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_PIPELINE | 214 ANV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK | 215 ANV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK)) { 216 uint32_t wm_depth_stencil_dw[GENX(3DSTATE_WM_DEPTH_STENCIL_length)]; 217 struct anv_dynamic_state *d = &cmd_buffer->state.dynamic; 218 219 struct GENX(3DSTATE_WM_DEPTH_STENCIL wm_depth_stencil) = { 220 GENX(3DSTATE_WM_DEPTH_STENCIL_header), 221 222 .StencilTestMask = d->stencil_compare_mask.front & 0xff, 223 .StencilWriteMask = d->stencil_write_mask.front & 0xff, 224 225 .BackfaceStencilTestMask = d->stencil_compare_mask.back & 0xff, 226 .BackfaceStencilWriteMask = d->stencil_write_mask.back & 0xff, 227 }; 228 GENX(3DSTATE_WM_DEPTH_STENCIL_pack)(NULL, wm_depth_stencil_dw, 229 &wm_depth_stencil); 230 231 anv_batch_emit_merge(&cmd_buffer->batch, wm_depth_stencil_dw, 232 pipeline->gen8.wm_depth_stencil); 233 } 234 #else 235 if (cmd_buffer->state.dirty & ANV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS) { 236 struct anv_state cc_state = 237 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, 238 GEN9_COLOR_CALC_STATE_length * 4, 239 64); 240 struct GEN9_COLOR_CALC_STATE cc = { 241 .BlendConstantColorRed = cmd_buffer->state.dynamic.blend_constants[0], 242 .BlendConstantColorGreen = cmd_buffer->state.dynamic.blend_constants[1], 243 .BlendConstantColorBlue = cmd_buffer->state.dynamic.blend_constants[2], 244 .BlendConstantColorAlpha = cmd_buffer->state.dynamic.blend_constants[3], 245 }; 246 GEN9_COLOR_CALC_STATE_pack(NULL, cc_state.map, &cc); 247 248 if (!cmd_buffer->device->info.has_llc) 249 anv_state_clflush(cc_state); 250 251 anv_batch_emit(&cmd_buffer->batch, GEN9_3DSTATE_CC_STATE_POINTERS, ccp) { 252 ccp.ColorCalcStatePointer = cc_state.offset; 253 ccp.ColorCalcStatePointerValid = true; 254 } 255 } 256 257 if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_PIPELINE | 258 ANV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK | 259 ANV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK | 260 ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE)) { 261 uint32_t dwords[GEN9_3DSTATE_WM_DEPTH_STENCIL_length]; 262 struct anv_dynamic_state *d = &cmd_buffer->state.dynamic; 263 struct GEN9_3DSTATE_WM_DEPTH_STENCIL wm_depth_stencil = { 264 GEN9_3DSTATE_WM_DEPTH_STENCIL_header, 265 266 .StencilTestMask = d->stencil_compare_mask.front & 0xff, 267 .StencilWriteMask = d->stencil_write_mask.front & 0xff, 268 269 .BackfaceStencilTestMask = d->stencil_compare_mask.back & 0xff, 270 .BackfaceStencilWriteMask = d->stencil_write_mask.back & 0xff, 271 272 .StencilReferenceValue = d->stencil_reference.front & 0xff, 273 .BackfaceStencilReferenceValue = d->stencil_reference.back & 0xff, 274 }; 275 GEN9_3DSTATE_WM_DEPTH_STENCIL_pack(NULL, dwords, &wm_depth_stencil); 276 277 anv_batch_emit_merge(&cmd_buffer->batch, dwords, 278 pipeline->gen9.wm_depth_stencil); 279 } 280 #endif 281 282 if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_PIPELINE | 283 ANV_CMD_DIRTY_INDEX_BUFFER)) { 284 anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_VF), vf) { 285 vf.IndexedDrawCutIndexEnable = pipeline->primitive_restart; 286 vf.CutIndex = cmd_buffer->state.restart_index; 287 } 288 } 289 290 cmd_buffer->state.dirty = 0; 291 } 292 293 void genX(CmdBindIndexBuffer)( 294 VkCommandBuffer commandBuffer, 295 VkBuffer _buffer, 296 VkDeviceSize offset, 297 VkIndexType indexType) 298 { 299 ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer); 300 ANV_FROM_HANDLE(anv_buffer, buffer, _buffer); 301 302 static const uint32_t vk_to_gen_index_type[] = { 303 [VK_INDEX_TYPE_UINT16] = INDEX_WORD, 304 [VK_INDEX_TYPE_UINT32] = INDEX_DWORD, 305 }; 306 307 static const uint32_t restart_index_for_type[] = { 308 [VK_INDEX_TYPE_UINT16] = UINT16_MAX, 309 [VK_INDEX_TYPE_UINT32] = UINT32_MAX, 310 }; 311 312 cmd_buffer->state.restart_index = restart_index_for_type[indexType]; 313 314 anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_INDEX_BUFFER), ib) { 315 ib.IndexFormat = vk_to_gen_index_type[indexType]; 316 ib.MemoryObjectControlState = GENX(MOCS); 317 ib.BufferStartingAddress = 318 (struct anv_address) { buffer->bo, buffer->offset + offset }; 319 ib.BufferSize = buffer->size - offset; 320 } 321 322 cmd_buffer->state.dirty |= ANV_CMD_DIRTY_INDEX_BUFFER; 323 } 324 325 /* Set of stage bits for which are pipelined, i.e. they get queued by the 326 * command streamer for later execution. 327 */ 328 #define ANV_PIPELINE_STAGE_PIPELINED_BITS \ 329 (VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | \ 330 VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | \ 331 VK_PIPELINE_STAGE_TESSELLATION_CONTROL_SHADER_BIT | \ 332 VK_PIPELINE_STAGE_TESSELLATION_EVALUATION_SHADER_BIT | \ 333 VK_PIPELINE_STAGE_GEOMETRY_SHADER_BIT | \ 334 VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | \ 335 VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | \ 336 VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT | \ 337 VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | \ 338 VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | \ 339 VK_PIPELINE_STAGE_TRANSFER_BIT | \ 340 VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT | \ 341 VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT | \ 342 VK_PIPELINE_STAGE_ALL_COMMANDS_BIT) 343 344 void genX(CmdSetEvent)( 345 VkCommandBuffer commandBuffer, 346 VkEvent _event, 347 VkPipelineStageFlags stageMask) 348 { 349 ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer); 350 ANV_FROM_HANDLE(anv_event, event, _event); 351 352 anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) { 353 if (stageMask & ANV_PIPELINE_STAGE_PIPELINED_BITS) { 354 pc.StallAtPixelScoreboard = true; 355 pc.CommandStreamerStallEnable = true; 356 } 357 358 pc.DestinationAddressType = DAT_PPGTT, 359 pc.PostSyncOperation = WriteImmediateData, 360 pc.Address = (struct anv_address) { 361 &cmd_buffer->device->dynamic_state_block_pool.bo, 362 event->state.offset 363 }; 364 pc.ImmediateData = VK_EVENT_SET; 365 } 366 } 367 368 void genX(CmdResetEvent)( 369 VkCommandBuffer commandBuffer, 370 VkEvent _event, 371 VkPipelineStageFlags stageMask) 372 { 373 ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer); 374 ANV_FROM_HANDLE(anv_event, event, _event); 375 376 anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) { 377 if (stageMask & ANV_PIPELINE_STAGE_PIPELINED_BITS) { 378 pc.StallAtPixelScoreboard = true; 379 pc.CommandStreamerStallEnable = true; 380 } 381 382 pc.DestinationAddressType = DAT_PPGTT; 383 pc.PostSyncOperation = WriteImmediateData; 384 pc.Address = (struct anv_address) { 385 &cmd_buffer->device->dynamic_state_block_pool.bo, 386 event->state.offset 387 }; 388 pc.ImmediateData = VK_EVENT_RESET; 389 } 390 } 391 392 void genX(CmdWaitEvents)( 393 VkCommandBuffer commandBuffer, 394 uint32_t eventCount, 395 const VkEvent* pEvents, 396 VkPipelineStageFlags srcStageMask, 397 VkPipelineStageFlags destStageMask, 398 uint32_t memoryBarrierCount, 399 const VkMemoryBarrier* pMemoryBarriers, 400 uint32_t bufferMemoryBarrierCount, 401 const VkBufferMemoryBarrier* pBufferMemoryBarriers, 402 uint32_t imageMemoryBarrierCount, 403 const VkImageMemoryBarrier* pImageMemoryBarriers) 404 { 405 ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer); 406 for (uint32_t i = 0; i < eventCount; i++) { 407 ANV_FROM_HANDLE(anv_event, event, pEvents[i]); 408 409 anv_batch_emit(&cmd_buffer->batch, GENX(MI_SEMAPHORE_WAIT), sem) { 410 sem.WaitMode = PollingMode, 411 sem.CompareOperation = COMPARE_SAD_EQUAL_SDD, 412 sem.SemaphoreDataDword = VK_EVENT_SET, 413 sem.SemaphoreAddress = (struct anv_address) { 414 &cmd_buffer->device->dynamic_state_block_pool.bo, 415 event->state.offset 416 }; 417 } 418 } 419 420 genX(CmdPipelineBarrier)(commandBuffer, srcStageMask, destStageMask, 421 false, /* byRegion */ 422 memoryBarrierCount, pMemoryBarriers, 423 bufferMemoryBarrierCount, pBufferMemoryBarriers, 424 imageMemoryBarrierCount, pImageMemoryBarriers); 425 } 426