1 /* 2 Copyright (C) Intel Corp. 2006. All Rights Reserved. 3 Intel funded Tungsten Graphics to 4 develop this 3D driver. 5 6 Permission is hereby granted, free of charge, to any person obtaining 7 a copy of this software and associated documentation files (the 8 "Software"), to deal in the Software without restriction, including 9 without limitation the rights to use, copy, modify, merge, publish, 10 distribute, sublicense, and/or sell copies of the Software, and to 11 permit persons to whom the Software is furnished to do so, subject to 12 the following conditions: 13 14 The above copyright notice and this permission notice (including the 15 next paragraph) shall be included in all copies or substantial 16 portions of the Software. 17 18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. 21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE 22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 25 26 **********************************************************************/ 27 /* 28 * Authors: 29 * Keith Whitwell <keithw (at) vmware.com> 30 */ 31 32 33 34 #include "intel_batchbuffer.h" 35 #include "intel_fbo.h" 36 #include "brw_context.h" 37 #include "brw_state.h" 38 #include "brw_defines.h" 39 #include "brw_wm.h" 40 #include "compiler/nir/nir.h" 41 42 /*********************************************************************** 43 * WM unit - fragment programs and rasterization 44 */ 45 46 bool 47 brw_color_buffer_write_enabled(struct brw_context *brw) 48 { 49 struct gl_context *ctx = &brw->ctx; 50 /* BRW_NEW_FRAGMENT_PROGRAM */ 51 const struct gl_program *fp = brw->fragment_program; 52 unsigned i; 53 54 /* _NEW_BUFFERS */ 55 for (i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; i++) { 56 struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[i]; 57 uint64_t outputs_written = fp->info.outputs_written; 58 59 /* _NEW_COLOR */ 60 if (rb && (outputs_written & BITFIELD64_BIT(FRAG_RESULT_COLOR) || 61 outputs_written & BITFIELD64_BIT(FRAG_RESULT_DATA0 + i)) && 62 (ctx->Color.ColorMask[i][0] || 63 ctx->Color.ColorMask[i][1] || 64 ctx->Color.ColorMask[i][2] || 65 ctx->Color.ColorMask[i][3])) { 66 return true; 67 } 68 } 69 70 return false; 71 } 72 73 /** 74 * Setup wm hardware state. See page 225 of Volume 2 75 */ 76 static void 77 brw_upload_wm_unit(struct brw_context *brw) 78 { 79 const struct gen_device_info *devinfo = &brw->screen->devinfo; 80 struct gl_context *ctx = &brw->ctx; 81 /* BRW_NEW_FRAGMENT_PROGRAM */ 82 const struct gl_program *fp = brw->fragment_program; 83 /* BRW_NEW_FS_PROG_DATA */ 84 const struct brw_wm_prog_data *prog_data = 85 brw_wm_prog_data(brw->wm.base.prog_data); 86 struct brw_wm_unit_state *wm; 87 88 wm = brw_state_batch(brw, AUB_TRACE_WM_STATE, 89 sizeof(*wm), 32, &brw->wm.base.state_offset); 90 memset(wm, 0, sizeof(*wm)); 91 92 if (prog_data->dispatch_8 && prog_data->dispatch_16) { 93 /* These two fields should be the same pre-gen6, which is why we 94 * only have one hardware field to program for both dispatch 95 * widths. 96 */ 97 assert(prog_data->base.dispatch_grf_start_reg == 98 prog_data->dispatch_grf_start_reg_2); 99 } 100 101 /* BRW_NEW_PROGRAM_CACHE | BRW_NEW_FS_PROG_DATA */ 102 wm->wm5.enable_8_pix = prog_data->dispatch_8; 103 wm->wm5.enable_16_pix = prog_data->dispatch_16; 104 105 if (prog_data->dispatch_8 || prog_data->dispatch_16) { 106 wm->thread0.grf_reg_count = prog_data->reg_blocks_0; 107 wm->thread0.kernel_start_pointer = 108 brw_program_reloc(brw, 109 brw->wm.base.state_offset + 110 offsetof(struct brw_wm_unit_state, thread0), 111 brw->wm.base.prog_offset + 112 (wm->thread0.grf_reg_count << 1)) >> 6; 113 } 114 115 if (prog_data->prog_offset_2) { 116 wm->wm9.grf_reg_count_2 = prog_data->reg_blocks_2; 117 wm->wm9.kernel_start_pointer_2 = 118 brw_program_reloc(brw, 119 brw->wm.base.state_offset + 120 offsetof(struct brw_wm_unit_state, wm9), 121 brw->wm.base.prog_offset + 122 prog_data->prog_offset_2 + 123 (wm->wm9.grf_reg_count_2 << 1)) >> 6; 124 } 125 126 wm->thread1.depth_coef_urb_read_offset = 1; 127 if (prog_data->base.use_alt_mode) 128 wm->thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754; 129 else 130 wm->thread1.floating_point_mode = BRW_FLOATING_POINT_IEEE_754; 131 132 wm->thread1.binding_table_entry_count = 133 prog_data->base.binding_table.size_bytes / 4; 134 135 if (prog_data->base.total_scratch != 0) { 136 wm->thread2.scratch_space_base_pointer = 137 brw->wm.base.scratch_bo->offset64 >> 10; /* reloc */ 138 wm->thread2.per_thread_scratch_space = 139 ffs(brw->wm.base.per_thread_scratch) - 11; 140 } else { 141 wm->thread2.scratch_space_base_pointer = 0; 142 wm->thread2.per_thread_scratch_space = 0; 143 } 144 145 wm->thread3.dispatch_grf_start_reg = 146 prog_data->base.dispatch_grf_start_reg; 147 wm->thread3.urb_entry_read_length = 148 prog_data->num_varying_inputs * 2; 149 wm->thread3.urb_entry_read_offset = 0; 150 wm->thread3.const_urb_entry_read_length = 151 prog_data->base.curb_read_length; 152 /* BRW_NEW_CURBE_OFFSETS */ 153 wm->thread3.const_urb_entry_read_offset = brw->curbe.wm_start * 2; 154 155 if (brw->gen == 5) 156 wm->wm4.sampler_count = 0; /* hardware requirement */ 157 else { 158 wm->wm4.sampler_count = (brw->wm.base.sampler_count + 1) / 4; 159 } 160 161 if (brw->wm.base.sampler_count) { 162 /* BRW_NEW_SAMPLER_STATE_TABLE - reloc */ 163 wm->wm4.sampler_state_pointer = (brw->batch.bo->offset64 + 164 brw->wm.base.sampler_offset) >> 5; 165 } else { 166 wm->wm4.sampler_state_pointer = 0; 167 } 168 169 /* BRW_NEW_FRAGMENT_PROGRAM */ 170 wm->wm5.program_uses_depth = prog_data->uses_src_depth; 171 wm->wm5.program_computes_depth = (fp->info.outputs_written & 172 BITFIELD64_BIT(FRAG_RESULT_DEPTH)) != 0; 173 /* _NEW_BUFFERS 174 * Override for NULL depthbuffer case, required by the Pixel Shader Computed 175 * Depth field. 176 */ 177 if (!intel_get_renderbuffer(ctx->DrawBuffer, BUFFER_DEPTH)) 178 wm->wm5.program_computes_depth = 0; 179 180 /* _NEW_COLOR */ 181 wm->wm5.program_uses_killpixel = 182 prog_data->uses_kill || ctx->Color.AlphaEnabled; 183 184 wm->wm5.max_threads = devinfo->max_wm_threads - 1; 185 186 /* _NEW_BUFFERS | _NEW_COLOR */ 187 if (brw_color_buffer_write_enabled(brw) || 188 wm->wm5.program_uses_killpixel || 189 wm->wm5.program_computes_depth) { 190 wm->wm5.thread_dispatch_enable = 1; 191 } 192 193 wm->wm5.legacy_line_rast = 0; 194 wm->wm5.legacy_global_depth_bias = 0; 195 wm->wm5.early_depth_test = 1; /* never need to disable */ 196 wm->wm5.line_aa_region_width = 0; 197 wm->wm5.line_endcap_aa_region_width = 1; 198 199 /* _NEW_POLYGONSTIPPLE */ 200 wm->wm5.polygon_stipple = ctx->Polygon.StippleFlag; 201 202 /* _NEW_POLYGON */ 203 if (ctx->Polygon.OffsetFill) { 204 wm->wm5.depth_offset = 1; 205 /* Something weird going on with legacy_global_depth_bias, 206 * offset_constant, scaling and MRD. This value passes glean 207 * but gives some odd results elsewere (eg. the 208 * quad-offset-units test). 209 */ 210 wm->global_depth_offset_constant = ctx->Polygon.OffsetUnits * 2; 211 212 /* This is the only value that passes glean: 213 */ 214 wm->global_depth_offset_scale = ctx->Polygon.OffsetFactor; 215 } 216 217 /* _NEW_LINE */ 218 wm->wm5.line_stipple = ctx->Line.StippleFlag; 219 220 /* BRW_NEW_STATS_WM */ 221 if (unlikely(INTEL_DEBUG & DEBUG_STATS) || brw->stats_wm) 222 wm->wm4.stats_enable = 1; 223 224 /* Emit scratch space relocation */ 225 if (prog_data->base.total_scratch != 0) { 226 drm_intel_bo_emit_reloc(brw->batch.bo, 227 brw->wm.base.state_offset + 228 offsetof(struct brw_wm_unit_state, thread2), 229 brw->wm.base.scratch_bo, 230 wm->thread2.per_thread_scratch_space, 231 I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER); 232 } 233 234 /* Emit sampler state relocation */ 235 if (brw->wm.base.sampler_count != 0) { 236 drm_intel_bo_emit_reloc(brw->batch.bo, 237 brw->wm.base.state_offset + 238 offsetof(struct brw_wm_unit_state, wm4), 239 brw->batch.bo, (brw->wm.base.sampler_offset | 240 wm->wm4.stats_enable | 241 (wm->wm4.sampler_count << 2)), 242 I915_GEM_DOMAIN_INSTRUCTION, 0); 243 } 244 245 brw->ctx.NewDriverState |= BRW_NEW_GEN4_UNIT_STATE; 246 247 /* _NEW_POLGYON */ 248 if (brw->wm.offset_clamp != ctx->Polygon.OffsetClamp) { 249 BEGIN_BATCH(2); 250 OUT_BATCH(_3DSTATE_GLOBAL_DEPTH_OFFSET_CLAMP << 16 | (2 - 2)); 251 OUT_BATCH_F(ctx->Polygon.OffsetClamp); 252 ADVANCE_BATCH(); 253 254 brw->wm.offset_clamp = ctx->Polygon.OffsetClamp; 255 } 256 } 257 258 const struct brw_tracked_state brw_wm_unit = { 259 .dirty = { 260 .mesa = _NEW_BUFFERS | 261 _NEW_COLOR | 262 _NEW_LINE | 263 _NEW_POLYGON | 264 _NEW_POLYGONSTIPPLE, 265 .brw = BRW_NEW_BATCH | 266 BRW_NEW_BLORP | 267 BRW_NEW_CURBE_OFFSETS | 268 BRW_NEW_FRAGMENT_PROGRAM | 269 BRW_NEW_FS_PROG_DATA | 270 BRW_NEW_PROGRAM_CACHE | 271 BRW_NEW_SAMPLER_STATE_TABLE | 272 BRW_NEW_STATS_WM, 273 }, 274 .emit = brw_upload_wm_unit, 275 }; 276
