src/mesa/drivers/dri/i965/gen7_gs_state.c

   1 /*
   2  * Copyright © 2013 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
  21  * DEALINGS IN THE SOFTWARE.
  22  */
  23
  24 #include "brw_context.h"
  25 #include "brw_state.h"
  26 #include "brw_defines.h"
  27 #include "intel_batchbuffer.h"
  28
  29
  30 static void
  31 gen7_upload_gs_push_constants(struct brw_context *brw)
  32 {
  33    /* BRW_NEW_GEOMETRY_PROGRAM */
  34    const struct brw_geometry_program *vp =
  35       (struct brw_geometry_program *) brw->geometry_program;
  36    if (!vp)
  37       return;
  38
  39    /* CACHE_NEW_GS_PROG */
  40    const struct brw_vec4_prog_data *prog_data = &brw->gs.prog_data->base;
  41    struct brw_stage_state *stage_state = &brw->gs.base;
  42
  43    gen6_upload_vec4_push_constants(brw, &vp->program.Base, prog_data,
  44                                    stage_state, AUB_TRACE_VS_CONSTANTS);
  45 }
  46
  47 const struct brw_tracked_state gen7_gs_push_constants = {
  48    .dirty = {
  49       .mesa  = _NEW_TRANSFORM | _NEW_PROGRAM_CONSTANTS,
  50       .brw   = (BRW_NEW_BATCH |
  51                 BRW_NEW_GEOMETRY_PROGRAM),
  52       .cache = CACHE_NEW_GS_PROG,
  53    },
  54    .emit = gen7_upload_gs_push_constants,
  55 };
  56
  57
  58 static void
  59 upload_gs_state(struct brw_context *brw)
  60 {
  61    const struct brw_stage_state *stage_state = &brw->gs.base;
  62    const int max_threads_shift = brw->is_haswell ?
  63       HSW_GS_MAX_THREADS_SHIFT : GEN6_GS_MAX_THREADS_SHIFT;
  64    /* BRW_NEW_GEOMETRY_PROGRAM */
  65    bool active = brw->geometry_program;
  66    /* CACHE_NEW_GS_PROG */
  67    const struct brw_vec4_prog_data *prog_data = &brw->gs.prog_data->base;
  68
  69    gen7_upload_constant_state(brw, stage_state, active, _3DSTATE_CONSTANT_GS);
  70
  71    /**
  72     * From Graphics BSpec: 3D-Media-GPGPU Engine > 3D Pipeline Stages >
  73     * Geometry > Geometry Shader > State:
  74     *
  75     *     "Note: Because of corruption in IVB:GT2, software needs to flush the
  76     *     whole fixed function pipeline when the GS enable changes value in
  77     *     the 3DSTATE_GS."
  78     *
  79     * The hardware architects have clarified that in this context "flush the
  80     * whole fixed function pipeline" means to emit a PIPE_CONTROL with the "CS
  81     * Stall" bit set.
  82     */
  83    if (!brw->is_haswell && brw->gt == 2 && brw->gs.enabled != active)
  84       gen7_emit_cs_stall_flush(brw);
  85
  86    if (active) {
  87       BEGIN_BATCH(7);
  88       OUT_BATCH(_3DSTATE_GS << 16 | (7 - 2));
  89       OUT_BATCH(stage_state->prog_offset);
  90       OUT_BATCH(((ALIGN(stage_state->sampler_count, 4)/4) <<
  91                  GEN6_GS_SAMPLER_COUNT_SHIFT) |
  92                 ((brw->gs.prog_data->base.base.binding_table.size_bytes / 4) <<
  93                  GEN6_GS_BINDING_TABLE_ENTRY_COUNT_SHIFT));
  94
  95       if (brw->gs.prog_data->base.total_scratch) {
  96          OUT_RELOC(stage_state->scratch_bo,
  97                    I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
  98                    ffs(brw->gs.prog_data->base.total_scratch) - 11);
  99       } else {
 100          OUT_BATCH(0);
 101       }
 102
 103       uint32_t dw4 =
 104          ((brw->gs.prog_data->output_vertex_size_hwords * 2 - 1) <<
 105           GEN7_GS_OUTPUT_VERTEX_SIZE_SHIFT) |
 106          (brw->gs.prog_data->output_topology <<
 107           GEN7_GS_OUTPUT_TOPOLOGY_SHIFT) |
 108          (prog_data->urb_read_length <<
 109           GEN6_GS_URB_READ_LENGTH_SHIFT) |
 110          (0 << GEN6_GS_URB_ENTRY_READ_OFFSET_SHIFT) |
 111          (prog_data->dispatch_grf_start_reg <<
 112           GEN6_GS_DISPATCH_START_GRF_SHIFT);
 113
 114       /* Note: the meaning of the GEN7_GS_REORDER_TRAILING bit changes between
 115        * Ivy Bridge and Haswell.
 116        *
 117        * On Ivy Bridge, setting this bit causes the vertices of a triangle
 118        * strip to be delivered to the geometry shader in an order that does
 119        * not strictly follow the OpenGL spec, but preserves triangle
 120        * orientation.  For example, if the vertices are (1, 2, 3, 4, 5), then
 121        * the geometry shader sees triangles:
 122        *
 123        * (1, 2, 3), (2, 4, 3), (3, 4, 5)
 124        *
 125        * (Clearing the bit is even worse, because it fails to preserve
 126        * orientation).
 127        *
 128        * Triangle strips with adjacency always ordered in a way that preserves
 129        * triangle orientation but does not strictly follow the OpenGL spec,
 130        * regardless of the setting of this bit.
 131        *
 132        * On Haswell, both triangle strips and triangle strips with adjacency
 133        * are always ordered in a way that preserves triangle orientation.
 134        * Setting this bit causes the ordering to strictly follow the OpenGL
 135        * spec.
 136        *
 137        * So in either case we want to set the bit.  Unfortunately on Ivy
 138        * Bridge this will get the order close to correct but not perfect.
 139        */
 140       uint32_t dw5 =
 141          ((brw->max_gs_threads - 1) << max_threads_shift) |
 142          (brw->gs.prog_data->control_data_header_size_hwords <<
 143           GEN7_GS_CONTROL_DATA_HEADER_SIZE_SHIFT) |
 144          ((brw->gs.prog_data->invocations - 1) <<
 145           GEN7_GS_INSTANCE_CONTROL_SHIFT) |
 146          (brw->gs.prog_data->dual_instanced_dispatch ?
 147           GEN7_GS_DISPATCH_MODE_DUAL_INSTANCE :
 148           GEN7_GS_DISPATCH_MODE_DUAL_OBJECT) |
 149          GEN6_GS_STATISTICS_ENABLE |
 150          (brw->gs.prog_data->include_primitive_id ?
 151           GEN7_GS_INCLUDE_PRIMITIVE_ID : 0) |
 152          GEN7_GS_REORDER_TRAILING |
 153          GEN7_GS_ENABLE;
 154       uint32_t dw6 = 0;
 155
 156       if (brw->is_haswell) {
 157          dw6 |= brw->gs.prog_data->control_data_format <<
 158             HSW_GS_CONTROL_DATA_FORMAT_SHIFT;
 159       } else {
 160          dw5 |= brw->gs.prog_data->control_data_format <<
 161             IVB_GS_CONTROL_DATA_FORMAT_SHIFT;
 162       }
 163
 164       OUT_BATCH(dw4);
 165       OUT_BATCH(dw5);
 166       OUT_BATCH(dw6);
 167       ADVANCE_BATCH();
 168    } else {
 169       BEGIN_BATCH(7);
 170       OUT_BATCH(_3DSTATE_GS << 16 | (7 - 2));
 171       OUT_BATCH(0); /* prog_bo */
 172       OUT_BATCH((0 << GEN6_GS_SAMPLER_COUNT_SHIFT) |
 173                 (0 << GEN6_GS_BINDING_TABLE_ENTRY_COUNT_SHIFT));
 174       OUT_BATCH(0); /* scratch space base offset */
 175       OUT_BATCH((1 << GEN6_GS_DISPATCH_START_GRF_SHIFT) |
 176                 (0 << GEN6_GS_URB_READ_LENGTH_SHIFT) |
 177                 GEN7_GS_INCLUDE_VERTEX_HANDLES |
 178                 (0 << GEN6_GS_URB_ENTRY_READ_OFFSET_SHIFT));
 179       OUT_BATCH((0 << GEN6_GS_MAX_THREADS_SHIFT) |
 180                 GEN6_GS_STATISTICS_ENABLE);
 181       OUT_BATCH(0);
 182       ADVANCE_BATCH();
 183    }
 184    brw->gs.enabled = active;
 185 }
 186
 187 const struct brw_tracked_state gen7_gs_state = {
 188    .dirty = {
 189       .mesa  = _NEW_TRANSFORM | _NEW_PROGRAM_CONSTANTS,
 190       .brw   = (BRW_NEW_CONTEXT |
 191                 BRW_NEW_GEOMETRY_PROGRAM |
 192                 BRW_NEW_BATCH |
 193                 BRW_NEW_PUSH_CONSTANT_ALLOCATION),
 194       .cache = CACHE_NEW_GS_PROG
 195    },
 196    .emit = upload_gs_state,
 197 };