Mesa (master): radeonsi: skip vs output optimizations for some outputs

[GitLab Mirror]

Module: Mesa
Branch: master
Commit: 17acff01a00109c87d59b9d876fc735dd5fbe3d1
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=17acff01a00109c87d59b9d876fc735dd5fbe3d1

Author: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer at amd.com>
Date:   Tue Apr 14 18:04:35 2020 +0200

radeonsi: skip vs output optimizations for some outputs

If PT_SPRITE_TEX is enabled, PS inputs are overriden at runtime so
we can't apply the vs output optim.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/2747
Fixes: 3ec9975555d ("radeonsi: eliminate trivial constant VS outputs")
Reviewed-by: Marek Olšák <marek.olsak at amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4559>

---

 src/amd/llvm/ac_llvm_build.c             |  4 ++++
 src/amd/llvm/ac_llvm_build.h             |  1 +
 src/amd/vulkan/radv_nir_to_llvm.c        |  2 +-
 src/gallium/drivers/radeonsi/si_shader.c | 13 ++++++++++++-
 4 files changed, 18 insertions(+), 2 deletions(-)

diff --git a/src/amd/llvm/ac_llvm_build.c b/src/amd/llvm/ac_llvm_build.c
index c2b5667f996..3d7589ee093 100644
--- a/src/amd/llvm/ac_llvm_build.c
+++ b/src/amd/llvm/ac_llvm_build.c
@@ -3081,6 +3081,7 @@ void ac_optimize_vs_outputs(struct ac_llvm_context *ctx,
 			    LLVMValueRef main_fn,
 			    uint8_t *vs_output_param_offset,
 			    uint32_t num_outputs,
+			    uint32_t skip_output_mask,
 			    uint8_t *num_param_exports)
 {
 	LLVMBasicBlockRef bb;
@@ -3124,6 +3125,9 @@ void ac_optimize_vs_outputs(struct ac_llvm_context *ctx,
 
 			target -= V_008DFC_SQ_EXP_PARAM;
 
+			if ((1u << target) & skip_output_mask)
+				continue;
+
 			/* Parse the instruction. */
 			memset(&exp, 0, sizeof(exp));
 			exp.offset = target;
diff --git a/src/amd/llvm/ac_llvm_build.h b/src/amd/llvm/ac_llvm_build.h
index ff90092754e..2c6142e76f9 100644
--- a/src/amd/llvm/ac_llvm_build.h
+++ b/src/amd/llvm/ac_llvm_build.h
@@ -611,6 +611,7 @@ void ac_optimize_vs_outputs(struct ac_llvm_context *ac,
 			    LLVMValueRef main_fn,
 			    uint8_t *vs_output_param_offset,
 			    uint32_t num_outputs,
+			    uint32_t skip_output_mask,
 			    uint8_t *num_param_exports);
 void ac_init_exec_full_mask(struct ac_llvm_context *ctx);
 
diff --git a/src/amd/vulkan/radv_nir_to_llvm.c b/src/amd/vulkan/radv_nir_to_llvm.c
index 85128ed0798..29a891627f7 100644
--- a/src/amd/vulkan/radv_nir_to_llvm.c
+++ b/src/amd/vulkan/radv_nir_to_llvm.c
@@ -3705,7 +3705,7 @@ ac_nir_eliminate_const_vs_outputs(struct radv_shader_context *ctx)
 	ac_optimize_vs_outputs(&ctx->ac,
 			       ctx->main_function,
 			       outinfo->vs_output_param_offset,
-			       VARYING_SLOT_MAX,
+			       VARYING_SLOT_MAX, 0,
 			       &outinfo->param_exports);
 }
 
diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index e615b81c293..7ae29880adb 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -1276,13 +1276,24 @@ static void si_optimize_vs_outputs(struct si_shader_context *ctx)
 {
    struct si_shader *shader = ctx->shader;
    struct si_shader_info *info = &shader->selector->info;
+   unsigned skip_vs_optim_mask = 0;
 
    if ((ctx->type != PIPE_SHADER_VERTEX && ctx->type != PIPE_SHADER_TESS_EVAL) ||
        shader->key.as_ls || shader->key.as_es)
       return;
 
+   /* Optimizing these outputs is not possible, since they might be overriden
+    * at runtime with S_028644_PT_SPRITE_TEX. */
+   for (int i = 0; i < info->num_outputs; i++) {
+      if (info->output_semantic_name[i] == TGSI_SEMANTIC_PCOORD ||
+          info->output_semantic_name[i] == TGSI_SEMANTIC_TEXCOORD) {
+         skip_vs_optim_mask |= 1u << shader->info.vs_output_param_offset[i];
+      }
+   }
+
    ac_optimize_vs_outputs(&ctx->ac, ctx->main_fn, shader->info.vs_output_param_offset,
-                          info->num_outputs, &shader->info.nr_param_exports);
+                          info->num_outputs, skip_vs_optim_mask,
+                          &shader->info.nr_param_exports);
 }
 
 static bool si_vs_needs_prolog(const struct si_shader_selector *sel,