src/compiler/nir/nir_lower_gs_intrinsics.c

   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 "nir.h"
  25 #include "nir_builder.h"
  26 #include "nir_xfb_info.h"
  27
  28 /**
  29  * \file nir_lower_gs_intrinsics.c
  30  *
  31  * Geometry Shaders can call EmitVertex()/EmitStreamVertex() to output an
  32  * arbitrary number of vertices.  However, the shader must declare the maximum
  33  * number of vertices that it will ever output - further attempts to emit
  34  * vertices result in undefined behavior according to the GLSL specification.
  35  *
  36  * Drivers might use this maximum number of vertices to allocate enough space
  37  * to hold the geometry shader's output.  Some drivers (such as i965) need to
  38  * implement "safety checks" which ensure that the shader hasn't emitted too
  39  * many vertices, to avoid overflowing that space and trashing other memory.
  40  *
  41  * The count of emitted vertices can also be useful in buffer offset
  42  * calculations, so drivers know where to write the GS output.
  43  *
  44  * However, for simple geometry shaders that emit a statically determinable
  45  * number of vertices, this extra bookkeeping is unnecessary and inefficient.
  46  * By tracking the vertex count in NIR, we allow constant folding/propagation
  47  * and dead control flow optimizations to eliminate most of it where possible.
  48  *
  49  * This pass introduces a new global variable which stores the current vertex
  50  * count (initialized to 0), and converts emit_vertex/end_primitive intrinsics
  51  * to their *_with_counter variants.  emit_vertex is also wrapped in a safety
  52  * check to avoid buffer overflows.  Finally, it adds a set_vertex_count
  53  * intrinsic at the end of the program, informing the driver of the final
  54  * vertex count.
  55  */
  56
  57 struct state {
  58    nir_builder *builder;
  59    nir_variable *vertex_count_vars[NIR_MAX_XFB_STREAMS];
  60    bool progress;
  61 };
  62
  63 /**
  64  * Replace emit_vertex intrinsics with:
  65  *
  66  * if (vertex_count < max_vertices) {
  67  *    emit_vertex_with_counter vertex_count ...
  68  *    vertex_count += 1
  69  * }
  70  */
  71 static void
  72 rewrite_emit_vertex(nir_intrinsic_instr *intrin, struct state *state)
  73 {
  74    nir_builder *b = state->builder;
  75    unsigned stream = nir_intrinsic_stream_id(intrin);
  76
  77    /* Load the vertex count */
  78    b->cursor = nir_before_instr(&intrin->instr);
  79    assert(state->vertex_count_vars[stream] != NULL);
  80    nir_ssa_def *count = nir_load_var(b, state->vertex_count_vars[stream]);
  81
  82    nir_ssa_def *max_vertices =
  83       nir_imm_int(b, b->shader->info.gs.vertices_out);
  84
  85    /* Create: if (vertex_count < max_vertices) and insert it.
  86     *
  87     * The new if statement needs to be hooked up to the control flow graph
  88     * before we start inserting instructions into it.
  89     */
  90    nir_push_if(b, nir_ilt(b, count, max_vertices));
  91
  92    nir_intrinsic_instr *lowered =
  93       nir_intrinsic_instr_create(b->shader,
  94                                  nir_intrinsic_emit_vertex_with_counter);
  95    nir_intrinsic_set_stream_id(lowered, stream);
  96    lowered->src[0] = nir_src_for_ssa(count);
  97    nir_builder_instr_insert(b, &lowered->instr);
  98
  99    /* Increment the vertex count by 1 */
 100    nir_store_var(b, state->vertex_count_vars[stream],
 101                  nir_iadd(b, count, nir_imm_int(b, 1)),
 102                  0x1); /* .x */
 103
 104    nir_pop_if(b, NULL);
 105
 106    nir_instr_remove(&intrin->instr);
 107
 108    state->progress = true;
 109 }
 110
 111 /**
 112  * Replace end_primitive with end_primitive_with_counter.
 113  */
 114 static void
 115 rewrite_end_primitive(nir_intrinsic_instr *intrin, struct state *state)
 116 {
 117    nir_builder *b = state->builder;
 118    unsigned stream = nir_intrinsic_stream_id(intrin);
 119
 120    b->cursor = nir_before_instr(&intrin->instr);
 121    assert(state->vertex_count_vars[stream] != NULL);
 122    nir_ssa_def *count = nir_load_var(b, state->vertex_count_vars[stream]);
 123
 124    nir_intrinsic_instr *lowered =
 125       nir_intrinsic_instr_create(b->shader,
 126                                  nir_intrinsic_end_primitive_with_counter);
 127    nir_intrinsic_set_stream_id(lowered, stream);
 128    lowered->src[0] = nir_src_for_ssa(count);
 129    nir_builder_instr_insert(b, &lowered->instr);
 130
 131    nir_instr_remove(&intrin->instr);
 132
 133    state->progress = true;
 134 }
 135
 136 static bool
 137 rewrite_intrinsics(nir_block *block, struct state *state)
 138 {
 139    nir_foreach_instr_safe(instr, block) {
 140       if (instr->type != nir_instr_type_intrinsic)
 141          continue;
 142
 143       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 144       switch (intrin->intrinsic) {
 145       case nir_intrinsic_emit_vertex:
 146          rewrite_emit_vertex(intrin, state);
 147          break;
 148       case nir_intrinsic_end_primitive:
 149          rewrite_end_primitive(intrin, state);
 150          break;
 151       default:
 152          /* not interesting; skip this */
 153          break;
 154       }
 155    }
 156
 157    return true;
 158 }
 159
 160 /**
 161  * Add a set_vertex_count intrinsic at the end of the program
 162  * (representing the final vertex count).
 163  */
 164 static void
 165 append_set_vertex_count(nir_block *end_block, struct state *state)
 166 {
 167    nir_builder *b = state->builder;
 168    nir_shader *shader = state->builder->shader;
 169
 170    /* Insert the new intrinsic in all of the predecessors of the end block,
 171     * but before any jump instructions (return).
 172     */
 173    set_foreach(end_block->predecessors, entry) {
 174       nir_block *pred = (nir_block *) entry->key;
 175       b->cursor = nir_after_block_before_jump(pred);
 176
 177       nir_ssa_def *count = nir_load_var(b, state->vertex_count_vars[0]);
 178
 179       nir_intrinsic_instr *set_vertex_count =
 180          nir_intrinsic_instr_create(shader, nir_intrinsic_set_vertex_count);
 181       set_vertex_count->src[0] = nir_src_for_ssa(count);
 182
 183       nir_builder_instr_insert(b, &set_vertex_count->instr);
 184    }
 185 }
 186
 187 bool
 188 nir_lower_gs_intrinsics(nir_shader *shader, bool per_stream)
 189 {
 190    struct state state;
 191    state.progress = false;
 192
 193    nir_function_impl *impl = nir_shader_get_entrypoint(shader);
 194    assert(impl);
 195
 196    nir_builder b;
 197    nir_builder_init(&b, impl);
 198    state.builder = &b;
 199
 200    /* Create the counter variables */
 201    b.cursor = nir_before_cf_list(&impl->body);
 202    for (unsigned i = 0; i < NIR_MAX_XFB_STREAMS; i++) {
 203       if (per_stream && !(shader->info.gs.active_stream_mask & (1 << i)))
 204          continue;
 205
 206       if (i == 0 || per_stream) {
 207          state.vertex_count_vars[i] =
 208             nir_local_variable_create(impl, glsl_uint_type(), "vertex_count");
 209          /* initialize to 0 */
 210          nir_store_var(&b, state.vertex_count_vars[i], nir_imm_int(&b, 0), 0x1);
 211       } else {
 212          /* If per_stream is false, we only have one counter which we want to use
 213           * for all streams.  Duplicate the counter pointer so all streams use the
 214           * same counter.
 215           */
 216          state.vertex_count_vars[i] = state.vertex_count_vars[0];
 217       }
 218    }
 219
 220    nir_foreach_block_safe(block, impl)
 221       rewrite_intrinsics(block, &state);
 222
 223    /* This only works because we have a single main() function. */
 224    if (!per_stream)
 225       append_set_vertex_count(impl->end_block, &state);
 226
 227    nir_metadata_preserve(impl, 0);
 228
 229    return state.progress;
 230 }