radeonsi: lower discard to demote when FS_CORRECT_DERIVS_AFTER_KILL is enabled

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4047>

diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index 01911cc8b54668971193a3009ee12170fa806e5c..8c8f90b80e23de98a0526bc4d8c76fe2c93baeb9 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -1599,13 +1599,6 @@ static bool si_build_main_function(struct si_shader_context *ctx,
                }
        }
 
-       if (sel->force_correct_derivs_after_kill) {
-               ctx->postponed_kill = ac_build_alloca_undef(&ctx->ac, ctx->ac.i1, "");
-               /* true = don't kill. */
-               LLVMBuildStore(ctx->ac.builder, ctx->ac.i1true,
-                              ctx->postponed_kill);
-       }
-
        bool success = si_nir_build_llvm(ctx, nir);
        if (free_nir)
                ralloc_free(nir);

diff --git a/src/gallium/drivers/radeonsi/si_shader.h b/src/gallium/drivers/radeonsi/si_shader.h
index 0109f68a9ed7f920bd34a626fd346e702961b053..1ee9aba9faf36ebba8e3858a99c683c920eb09de 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader.h
+++ b/src/gallium/drivers/radeonsi/si_shader.h
@@ -435,7 +435,6 @@ struct si_shader_selector {
        /* PIPE_SHADER_[VERTEX|FRAGMENT|...] */
        enum pipe_shader_type type;
        bool            vs_needs_prolog;
-       bool            force_correct_derivs_after_kill;
        bool            prim_discard_cs_allowed;
        bool            ngg_culling_allowed;
        unsigned        num_vs_inputs;

diff --git a/src/gallium/drivers/radeonsi/si_shader_llvm_ps.c b/src/gallium/drivers/radeonsi/si_shader_llvm_ps.c
index 493eb2cf908a6942dbfc09e0b666aad4e25d0574..ea8077dcc2800f2d2c8daef2a7d08fb8d4917da9 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader_llvm_ps.c
+++ b/src/gallium/drivers/radeonsi/si_shader_llvm_ps.c
@@ -621,17 +621,6 @@ static void si_llvm_emit_kill(struct ac_shader_abi *abi, LLVMValueRef visible)
        struct si_shader_context *ctx = si_shader_context_from_abi(abi);
        LLVMBuilderRef builder = ctx->ac.builder;
 
-       if (ctx->shader->selector->force_correct_derivs_after_kill) {
-               /* Kill immediately while maintaining WQM. */
-               ac_build_kill_if_false(&ctx->ac,
-                                      ac_build_wqm_vote(&ctx->ac, visible));
-
-               LLVMValueRef mask = LLVMBuildLoad(builder, ctx->postponed_kill, "");
-               mask = LLVMBuildAnd(builder, mask, visible, "");
-               LLVMBuildStore(builder, mask, ctx->postponed_kill);
-               return;
-       }
-
        ac_build_kill_if_false(&ctx->ac, visible);
 }
 

diff --git a/src/gallium/drivers/radeonsi/si_shader_nir.c b/src/gallium/drivers/radeonsi/si_shader_nir.c
index 812683e28812129ab1108f2131eb64771aa7b723..da97b738c40ac8be82c700969b00eb646fcc5c10 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader_nir.c
+++ b/src/gallium/drivers/radeonsi/si_shader_nir.c
@@ -985,6 +985,9 @@ static void si_lower_nir(struct si_screen *sscreen, struct nir_shader *nir)
 
        NIR_PASS_V(nir, nir_lower_bool_to_int32);
        NIR_PASS_V(nir, nir_remove_dead_variables, nir_var_function_temp);
+
+       if (sscreen->debug_flags & DBG(FS_CORRECT_DERIVS_AFTER_KILL))
+               NIR_PASS_V(nir, nir_lower_discard_to_demote);
 }
 
 void si_finalize_nir(struct pipe_screen *screen, void *nirptr, bool optimize)

diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c
index fcf99088164db333caa5e34b392fe05c6700c740..ebb3fa30b8807ac1b08e08b8f936b1568b7befeb 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -73,7 +73,10 @@ void si_get_ir_cache_key(struct si_shader_selector *sel, bool ngg, bool es,
                shader_variant_flags |= 1 << 1;
        if (si_get_wave_size(sel->screen, sel->type, ngg, es) == 32)
                shader_variant_flags |= 1 << 2;
-       if (sel->force_correct_derivs_after_kill)
+       if (sel->type == PIPE_SHADER_FRAGMENT &&
+           sel->info.uses_derivatives &&
+           sel->info.uses_kill &&
+           sel->screen->debug_flags & DBG(FS_CORRECT_DERIVS_AFTER_KILL))
                shader_variant_flags |= 1 << 3;
 
        struct mesa_sha1 ctx;
@@ -2822,12 +2825,6 @@ static void *si_create_shader_selector(struct pipe_context *ctx,
                               sel->info.num_inputs &&
                               !sel->info.properties[TGSI_PROPERTY_VS_BLIT_SGPRS_AMD];
 
-       sel->force_correct_derivs_after_kill =
-               sel->type == PIPE_SHADER_FRAGMENT &&
-               sel->info.uses_derivatives &&
-               sel->info.uses_kill &&
-               sctx->screen->debug_flags & DBG(FS_CORRECT_DERIVS_AFTER_KILL);
-
        sel->prim_discard_cs_allowed =
                sel->type == PIPE_SHADER_VERTEX &&
                !sel->info.uses_bindless_images &&