aco, radv: Move PS epilog and VS prolog args to their info structs.

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21696>

src/amd/compiler/aco_instruction_selection.cpp

@@ -11114,7 +11114,7 @@ create_fs_jump_to_epilog(isel_context* ctx)
       }
    }
 
-   Temp continue_pc = convert_pointer_to_64_bit(ctx, get_arg(ctx, ctx->args->ps_epilog_pc));
+   Temp continue_pc = convert_pointer_to_64_bit(ctx, get_arg(ctx, ctx->options->key.ps.epilog.pc));
 
    aco_ptr<Pseudo_instruction> jump{create_instruction<Pseudo_instruction>(
       aco_opcode::p_jump_to_epilog, Format::PSEUDO, 1 + color_exports.size(), 0)};
@@ -11797,12 +11797,13 @@ load_vb_descs(Builder& bld, PhysReg dest, Operand base, unsigned start, unsigned
 }
 
 Operand
-calc_nontrivial_instance_id(Builder& bld, const struct radv_shader_args* args, unsigned index,
+calc_nontrivial_instance_id(Builder& bld, const struct radv_shader_args* args,
+                            const struct aco_vs_prolog_info* pinfo, unsigned index,
                             Operand instance_id, Operand start_instance, PhysReg tmp_sgpr,
                             PhysReg tmp_vgpr0, PhysReg tmp_vgpr1)
 {
    bld.smem(aco_opcode::s_load_dwordx2, Definition(tmp_sgpr, s2),
-            get_arg_fixed(args, args->prolog_inputs), Operand::c32(8u + index * 8u));
+            get_arg_fixed(args, pinfo->inputs), Operand::c32(8u + index * 8u));
 
    wait_imm lgkm_imm;
    lgkm_imm.lgkm = 0;
@@ -11977,7 +11978,7 @@ select_vs_prolog(Program* program, const struct aco_vs_prolog_info* pinfo, ac_sh
                   unsigned index =
                      util_bitcount(pinfo->state.nontrivial_divisors & BITFIELD_MASK(loc));
                   fetch_index = calc_nontrivial_instance_id(
-                     bld, args, index, instance_id, start_instance, prolog_input,
+                     bld, args, pinfo, index, instance_id, start_instance, prolog_input,
                      nontrivial_tmp_vgpr0, nontrivial_tmp_vgpr1);
                } else {
                   fetch_index = Operand(instance_index, v1);
@@ -12083,10 +12084,10 @@ select_vs_prolog(Program* program, const struct aco_vs_prolog_info* pinfo, ac_sh
    block->kind |= block_kind_uniform;
 
    /* continue on to the main shader */
-   Operand continue_pc = get_arg_fixed(args, args->prolog_inputs);
+   Operand continue_pc = get_arg_fixed(args, pinfo->inputs);
    if (has_nontrivial_divisors) {
       bld.smem(aco_opcode::s_load_dwordx2, Definition(prolog_input, s2),
-               get_arg_fixed(args, args->prolog_inputs), Operand::c32(0u));
+               get_arg_fixed(args, pinfo->inputs), Operand::c32(0u));
       bld.sopp(aco_opcode::s_waitcnt, -1, lgkm_imm.pack(program->gfx_level));
       continue_pc = Operand(prolog_input, s2);
    }
@@ -12133,7 +12134,7 @@ select_ps_epilog(Program* program, const struct aco_ps_epilog_info* einfo, ac_sh
       out.is_int10 = (einfo->color_is_int10 >> i) & 1;
       out.enable_mrt_output_nan_fixup = (einfo->enable_mrt_output_nan_fixup >> i) & 1;
 
-      Temp inputs = get_arg(&ctx, ctx.args->ps_epilog_inputs[i]);
+      Temp inputs = get_arg(&ctx, einfo->inputs[i]);
       for (unsigned c = 0; c < 4; ++c) {
          out.values[c] = Operand(emit_extract_vector(&ctx, inputs, c, v1));
       }

src/amd/compiler/aco_shader_info.h

@@ -27,6 +27,7 @@
 #ifndef ACO_SHADER_INFO_H
 #define ACO_SHADER_INFO_H
 
+#include "ac_shader_args.h"
 #include "shader_enums.h"
 
 #ifdef __cplusplus
@@ -53,6 +54,7 @@ struct aco_vs_input_state {
 };
 
 struct aco_vs_prolog_info {
+   struct ac_arg inputs;
    struct aco_vs_input_state state;
    unsigned num_attributes;
    uint32_t misaligned_mask;
@@ -132,6 +134,9 @@ enum aco_compiler_debug_level {
 };
 
 struct aco_ps_epilog_info {
+   struct ac_arg inputs[8];
+   struct ac_arg pc;
+
    uint32_t spi_shader_col_format;
 
    /* Bitmasks, each bit represents one of the 8 MRTs. */

src/amd/vulkan/radv_aco_shader_info.h

@@ -97,7 +97,8 @@ radv_aco_convert_shader_info(struct aco_shader_info *aco_info,
 #define ASSIGN_VS_STATE_FIELD_CP(x) memcpy(&aco_info->state.x, &radv->state->x, sizeof(radv->state->x))
 static inline void
 radv_aco_convert_vs_prolog_key(struct aco_vs_prolog_info *aco_info,
-                               const struct radv_vs_prolog_key *radv)
+                               const struct radv_vs_prolog_key *radv,
+                               const struct radv_shader_args *radv_args)
 {
    ASSIGN_VS_STATE_FIELD(instance_rate_inputs);
    ASSIGN_VS_STATE_FIELD(nontrivial_divisors);
@@ -110,24 +111,30 @@ radv_aco_convert_vs_prolog_key(struct aco_vs_prolog_info *aco_info,
    ASSIGN_FIELD(misaligned_mask);
    ASSIGN_FIELD(is_ngg);
    ASSIGN_FIELD(next_stage);
+
+   aco_info->inputs = radv_args->prolog_inputs;
 }
 
 static inline void
 radv_aco_convert_ps_epilog_key(struct aco_ps_epilog_info *aco_info,
-                               const struct radv_ps_epilog_key *radv)
+                               const struct radv_ps_epilog_key *radv,
+                               const struct radv_shader_args *radv_args)
 {
    ASSIGN_FIELD(spi_shader_col_format);
    ASSIGN_FIELD(color_is_int8);
    ASSIGN_FIELD(color_is_int10);
    ASSIGN_FIELD(enable_mrt_output_nan_fixup);
    ASSIGN_FIELD(mrt0_is_dual_src);
+
+   memcpy(aco_info->inputs, radv_args->ps_epilog_inputs, sizeof(aco_info->inputs));
+   aco_info->pc = radv_args->ps_epilog_pc;
 }
 
 static inline void
-radv_aco_convert_pipe_key(struct aco_stage_input *aco_info,
-                          const struct radv_pipeline_key *radv)
+radv_aco_convert_pipe_key(struct aco_stage_input *aco_info, const struct radv_pipeline_key *radv,
+                          const struct radv_shader_args *radv_args)
 {
-   radv_aco_convert_ps_epilog_key(&aco_info->ps.epilog, &radv->ps.epilog);
+   radv_aco_convert_ps_epilog_key(&aco_info->ps.epilog, &radv->ps.epilog, radv_args);
    ASSIGN_FIELD(optimisations_disabled);
    ASSIGN_FIELD(image_2d_view_of_3d);
    ASSIGN_FIELD(vs.instance_rate_inputs);
@@ -146,7 +153,7 @@ radv_aco_convert_opts(struct aco_compiler_options *aco_info,
                       const struct radv_nir_compiler_options *radv,
                       const struct radv_shader_args *radv_args)
 {
-   radv_aco_convert_pipe_key(&aco_info->key, &radv->key);
+   radv_aco_convert_pipe_key(&aco_info->key, &radv->key, radv_args);
    ASSIGN_FIELD(robust_buffer_access);
    ASSIGN_FIELD(dump_shader);
    ASSIGN_FIELD(dump_preoptir);

src/amd/vulkan/radv_shader.c

@@ -2544,7 +2544,7 @@ radv_create_vs_prolog(struct radv_device *device, const struct radv_vs_prolog_ke
    struct aco_compiler_options ac_opts;
    radv_aco_convert_shader_info(&ac_info, &info, &args);
    radv_aco_convert_opts(&ac_opts, &options, &args);
-   radv_aco_convert_vs_prolog_key(&ac_prolog_info, key);
+   radv_aco_convert_vs_prolog_key(&ac_prolog_info, key, &args);
    aco_compile_vs_prolog(&ac_opts, &ac_info, &ac_prolog_info, &args, &radv_aco_build_shader_part,
                          (void **)&binary);
 
@@ -2609,7 +2609,7 @@ radv_create_ps_epilog(struct radv_device *device, const struct radv_ps_epilog_ke
    struct aco_compiler_options ac_opts;
    radv_aco_convert_shader_info(&ac_info, &info, &args);
    radv_aco_convert_opts(&ac_opts, &options, &args);
-   radv_aco_convert_ps_epilog_key(&ac_epilog_info, key);
+   radv_aco_convert_ps_epilog_key(&ac_epilog_info, key, &args);
    aco_compile_ps_epilog(&ac_opts, &ac_info, &ac_epilog_info, &args, &radv_aco_build_shader_part,
                          (void **)&binary);