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    nir_ssa_def *count = nir_load_var(b, state->vertex_count_vars[stream]);
  80
  81    nir_ssa_def *max_vertices =
  82       nir_imm_int(b, b->shader->info.gs.vertices_out);
  83
  84    /* Create: if (vertex_count < max_vertices) and insert it.
  85     *
  86     * The new if statement needs to be hooked up to the control flow graph
  87     * before we start inserting instructions into it.
  88     */
  89    nir_push_if(b, nir_ilt(b, count, max_vertices));
  90
  91    nir_intrinsic_instr *lowered =
  92       nir_intrinsic_instr_create(b->shader,
  93                                  nir_intrinsic_emit_vertex_with_counter);
  94    nir_intrinsic_set_stream_id(lowered, stream);
  95    lowered->src[0] = nir_src_for_ssa(count);
  96    nir_builder_instr_insert(b, &lowered->instr);
  97
  98    /* Increment the vertex count by 1 */
  99    nir_store_var(b, state->vertex_count_vars[stream],
 100                  nir_iadd(b, count, nir_imm_int(b, 1)),
 101                  0x1); /* .x */
 102
 103    nir_pop_if(b, NULL);
 104
 105    nir_instr_remove(&intrin->instr);
 106
 107    state->progress = true;
 108 }
 109
 110 /**
 111  * Replace end_primitive with end_primitive_with_counter.
 112  */
 113 static void
 114 rewrite_end_primitive(nir_intrinsic_instr *intrin, struct state *state)
 115 {
 116    nir_builder *b = state->builder;
 117    unsigned stream = nir_intrinsic_stream_id(intrin);
 118
 119    b->cursor = nir_before_instr(&intrin->instr);
 120    nir_ssa_def *count = nir_load_var(b, state->vertex_count_vars[stream]);
 121
 122    nir_intrinsic_instr *lowered =
 123       nir_intrinsic_instr_create(b->shader,
 124                                  nir_intrinsic_end_primitive_with_counter);
 125    nir_intrinsic_set_stream_id(lowered, stream);
 126    lowered->src[0] = nir_src_for_ssa(count);
 127    nir_builder_instr_insert(b, &lowered->instr);
 128
 129    nir_instr_remove(&intrin->instr);
 130
 131    state->progress = true;
 132 }
 133
 134 static bool
 135 rewrite_intrinsics(nir_block *block, struct state *state)
 136 {
 137    nir_foreach_instr_safe(instr, block) {
 138       if (instr->type != nir_instr_type_intrinsic)
 139          continue;
 140
 141       nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 142       switch (intrin->intrinsic) {
 143       case nir_intrinsic_emit_vertex:
 144          rewrite_emit_vertex(intrin, state);
 145          break;
 146       case nir_intrinsic_end_primitive:
 147          rewrite_end_primitive(intrin, state);
 148          break;
 149       default:
 150          /* not interesting; skip this */
 151          break;
 152       }
 153    }
 154
 155    return true;
 156 }
 157
 158 /**
 159  * Add a set_vertex_count intrinsic at the end of the program
 160  * (representing the final vertex count).
 161  */
 162 static void
 163 append_set_vertex_count(nir_block *end_block, struct state *state)
 164 {
 165    nir_builder *b = state->builder;
 166    nir_shader *shader = state->builder->shader;
 167
 168    /* Insert the new intrinsic in all of the predecessors of the end block,
 169     * but before any jump instructions (return).
 170     */
 171    set_foreach(end_block->predecessors, entry) {
 172       nir_block *pred = (nir_block *) entry->key;
 173       b->cursor = nir_after_block_before_jump(pred);
 174
 175       nir_ssa_def *count = nir_load_var(b, state->vertex_count_vars[0]);
 176
 177       nir_intrinsic_instr *set_vertex_count =
 178          nir_intrinsic_instr_create(shader, nir_intrinsic_set_vertex_count);
 179       set_vertex_count->src[0] = nir_src_for_ssa(count);
 180
 181       nir_builder_instr_insert(b, &set_vertex_count->instr);
 182    }
 183 }
 184
 185 bool
 186 nir_lower_gs_intrinsics(nir_shader *shader, bool per_stream)
 187 {
 188    struct state state;
 189    state.progress = false;
 190
 191    nir_function_impl *impl = nir_shader_get_entrypoint(shader);
 192    assert(impl);
 193
 194    nir_builder b;
 195    nir_builder_init(&b, impl);
 196    state.builder = &b;
 197
 198    /* Create the counter variables */
 199    b.cursor = nir_before_cf_list(&impl->body);
 200    unsigned num_counters = per_stream && shader->info.gs.uses_streams ?
 201                            NIR_MAX_XFB_STREAMS : 1;
 202    for (unsigned i = 0; i < num_counters; i++) {
 203       state.vertex_count_vars[i] =
 204          nir_local_variable_create(impl, glsl_uint_type(), "vertex_count");
 205       /* initialize to 0 */
 206       nir_store_var(&b, state.vertex_count_vars[i], nir_imm_int(&b, 0), 0x1);
 207    }
 208    /* If per_stream is false, we only have one counter which we want to use
 209     * for all streams.  Duplicate the counter pointer so all streams use the
 210     * same counter.
 211     */
 212    for (unsigned i = num_counters; i < NIR_MAX_XFB_STREAMS; i++)
 213       state.vertex_count_vars[i] = state.vertex_count_vars[0];
 214
 215    nir_foreach_block_safe(block, impl)
 216       rewrite_intrinsics(block, &state);
 217
 218    /* This only works because we have a single main() function. */
 219    if (!per_stream)
 220       append_set_vertex_count(impl->end_block, &state);
 221
 222    nir_metadata_preserve(impl, 0);
 223
 224    return state.progress;
 225 }