anv: Use a dynamic array for storing executables in pipeline

Avoids waste for pipelines that don't use all the shaders, and is
flexible enough to cover cases where there are multiple variants per
shader (e.g. SIMD8/16/32 for fragment shader).

Even though we could pre-calculate the exact size of the array, this
is not a critical path so it is worth preventing the bug that will
likely happen when new variants are added but not accounted for.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4040>

diff --git a/src/intel/vulkan/anv_pipeline.c b/src/intel/vulkan/anv_pipeline.c
index eeeb9f324ed5d89bb180f20ea36b7f5856c716a6..9f14fb88f821d7b2727cd0e570f13af41ab743ee 100644 (file)
--- a/src/intel/vulkan/anv_pipeline.c
+++ b/src/intel/vulkan/anv_pipeline.c
@@ -1051,13 +1051,14 @@ anv_pipeline_add_executable(struct anv_pipeline *pipeline,
       free(stream_data);
    }
 
-   pipeline->executables[pipeline->num_executables++] =
-      (struct anv_pipeline_executable) {
-         .stage = stage->stage,
-         .stats = *stats,
-         .nir = nir,
-         .disasm = disasm,
-      };
+   const struct anv_pipeline_executable exe = {
+      .stage = stage->stage,
+      .stats = *stats,
+      .nir = nir,
+      .disasm = disasm,
+   };
+   util_dynarray_append(&pipeline->executables,
+                        struct anv_pipeline_executable, exe);
 }
 
 static void
@@ -1873,7 +1874,8 @@ anv_pipeline_init(struct anv_pipeline *pipeline,
     * of various prog_data pointers.  Make them NULL by default.
     */
    memset(pipeline->shaders, 0, sizeof(pipeline->shaders));
-   pipeline->num_executables = 0;
+
+   util_dynarray_init(&pipeline->executables, pipeline->mem_ctx);
 
    result = anv_pipeline_compile_graphics(pipeline, cache, pCreateInfo);
    if (result != VK_SUCCESS) {
@@ -1976,12 +1978,12 @@ VkResult anv_GetPipelineExecutablePropertiesKHR(
    ANV_FROM_HANDLE(anv_pipeline, pipeline, pPipelineInfo->pipeline);
    VK_OUTARRAY_MAKE(out, pProperties, pExecutableCount);
 
-   for (uint32_t i = 0; i < pipeline->num_executables; i++) {
+   util_dynarray_foreach (&pipeline->executables, struct anv_pipeline_executable, exe) {
       vk_outarray_append(&out, props) {
-         gl_shader_stage stage = pipeline->executables[i].stage;
+         gl_shader_stage stage = exe->stage;
          props->stages = mesa_to_vk_shader_stage(stage);
 
-         unsigned simd_width = pipeline->executables[i].stats.dispatch_width;
+         unsigned simd_width = exe->stats.dispatch_width;
          if (stage == MESA_SHADER_FRAGMENT) {
             WRITE_STR(props->name, "%s%d %s",
                       simd_width ? "SIMD" : "vec",
@@ -2005,6 +2007,15 @@ VkResult anv_GetPipelineExecutablePropertiesKHR(
    return vk_outarray_status(&out);
 }
 
+static const struct anv_pipeline_executable *
+anv_pipeline_get_executable(struct anv_pipeline *pipeline, uint32_t index)
+{
+   assert(index < util_dynarray_num_elements(&pipeline->executables,
+                                             struct anv_pipeline_executable));
+   return util_dynarray_element(
+      &pipeline->executables, struct anv_pipeline_executable, index);
+}
+
 VkResult anv_GetPipelineExecutableStatisticsKHR(
     VkDevice                                    device,
     const VkPipelineExecutableInfoKHR*          pExecutableInfo,
@@ -2014,9 +2025,8 @@ VkResult anv_GetPipelineExecutableStatisticsKHR(
    ANV_FROM_HANDLE(anv_pipeline, pipeline, pExecutableInfo->pipeline);
    VK_OUTARRAY_MAKE(out, pStatistics, pStatisticCount);
 
-   assert(pExecutableInfo->executableIndex < pipeline->num_executables);
    const struct anv_pipeline_executable *exe =
-      &pipeline->executables[pExecutableInfo->executableIndex];
+      anv_pipeline_get_executable(pipeline, pExecutableInfo->executableIndex);
    const struct brw_stage_prog_data *prog_data =
       pipeline->shaders[exe->stage]->prog_data;
 
@@ -2127,9 +2137,8 @@ VkResult anv_GetPipelineExecutableInternalRepresentationsKHR(
                     pInternalRepresentationCount);
    bool incomplete_text = false;
 
-   assert(pExecutableInfo->executableIndex < pipeline->num_executables);
    const struct anv_pipeline_executable *exe =
-      &pipeline->executables[pExecutableInfo->executableIndex];
+      anv_pipeline_get_executable(pipeline, pExecutableInfo->executableIndex);
 
    if (exe->nir) {
       vk_outarray_append(&out, ir) {

diff --git a/src/intel/vulkan/anv_private.h b/src/intel/vulkan/anv_private.h
index cdb2a725a91ba93a6b624c367189589e054f0138..3607b7d4fcd80f29722b1701fa3c5e875bd1a620 100644 (file)
--- a/src/intel/vulkan/anv_private.h
+++ b/src/intel/vulkan/anv_private.h
@@ -3145,9 +3145,6 @@ anv_shader_bin_unref(struct anv_device *device, struct anv_shader_bin *shader)
       anv_shader_bin_destroy(device, shader);
 }
 
-/* 5 possible simultaneous shader stages and FS may have up to 3 binaries */
-#define MAX_PIPELINE_EXECUTABLES 7
-
 struct anv_pipeline_executable {
    gl_shader_stage stage;
 
@@ -3178,8 +3175,7 @@ struct anv_pipeline {
 
    struct anv_shader_bin *                      shaders[MESA_SHADER_STAGES];
 
-   uint32_t                                     num_executables;
-   struct anv_pipeline_executable               executables[MAX_PIPELINE_EXECUTABLES];
+   struct util_dynarray                         executables;
 
    const struct gen_l3_config *                 l3_config;
 

diff --git a/src/intel/vulkan/genX_pipeline.c b/src/intel/vulkan/genX_pipeline.c
index c5c5e73874ec8128a8707837872411d8117cb76b..5b692a7e8130fad26172543584a9fa4500e917bc 100644 (file)
--- a/src/intel/vulkan/genX_pipeline.c
+++ b/src/intel/vulkan/genX_pipeline.c
@@ -2256,7 +2256,8 @@ compute_pipeline_create(
     * of various prog_data pointers.  Make them NULL by default.
     */
    memset(pipeline->shaders, 0, sizeof(pipeline->shaders));
-   pipeline->num_executables = 0;
+
+   util_dynarray_init(&pipeline->executables, pipeline->mem_ctx);
 
    assert(pCreateInfo->stage.stage == VK_SHADER_STAGE_COMPUTE_BIT);
    pipeline->active_stages |= VK_SHADER_STAGE_COMPUTE_BIT;