vkd3d-shader/tpf: Introduce tpf_write_shader_function().

This commit is contained in:
Shaun Ren 2024-10-15 16:33:21 -04:00 committed by Henri Verbeet
parent 147f8898ca
commit df3aaac7cf
Notes: Henri Verbeet 2024-10-16 21:47:54 +02:00
Approved-by: Elizabeth Figura (@zfigura)
Approved-by: Henri Verbeet (@hverbeet)
Merge-Request: https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/1187
3 changed files with 53 additions and 43 deletions

View File

@ -1429,6 +1429,7 @@ void hlsl_lower_index_loads(struct hlsl_ctx *ctx, struct hlsl_block *body);
void hlsl_run_const_passes(struct hlsl_ctx *ctx, struct hlsl_block *body); void hlsl_run_const_passes(struct hlsl_ctx *ctx, struct hlsl_block *body);
uint32_t allocate_temp_registers(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry_func); uint32_t allocate_temp_registers(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry_func);
void mark_indexable_vars(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry_func); void mark_indexable_vars(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry_func);
void compute_liveness(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry_func);
int hlsl_emit_bytecode(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry_func, int hlsl_emit_bytecode(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry_func,
enum vkd3d_shader_target_type target_type, struct vkd3d_shader_code *out); enum vkd3d_shader_target_type target_type, struct vkd3d_shader_code *out);
int hlsl_emit_effect_binary(struct hlsl_ctx *ctx, struct vkd3d_shader_code *out); int hlsl_emit_effect_binary(struct hlsl_ctx *ctx, struct vkd3d_shader_code *out);

View File

@ -4375,7 +4375,7 @@ static void init_var_liveness(struct hlsl_ir_var *var)
var->last_read = UINT_MAX; var->last_read = UINT_MAX;
} }
static void compute_liveness(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry_func) void compute_liveness(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry_func)
{ {
struct hlsl_scope *scope; struct hlsl_scope *scope;
struct hlsl_ir_var *var; struct hlsl_ir_var *var;

View File

@ -6496,6 +6496,56 @@ static void write_sm4_block(const struct tpf_compiler *tpf, const struct hlsl_bl
} }
} }
static void tpf_write_shader_function(struct tpf_compiler *tpf, struct hlsl_ir_function_decl *func)
{
struct hlsl_ctx *ctx = tpf->ctx;
const struct hlsl_scope *scope;
const struct hlsl_ir_var *var;
uint32_t temp_count;
compute_liveness(ctx, func);
mark_indexable_vars(ctx, func);
temp_count = allocate_temp_registers(ctx, func);
if (ctx->result)
return;
LIST_FOR_EACH_ENTRY(var, &func->extern_vars, struct hlsl_ir_var, extern_entry)
{
if ((var->is_input_semantic && var->last_read)
|| (var->is_output_semantic && var->first_write))
write_sm4_dcl_semantic(tpf, var);
}
if (tpf->program->shader_version.type == VKD3D_SHADER_TYPE_COMPUTE)
write_sm4_dcl_thread_group(tpf, ctx->thread_count);
if (temp_count)
write_sm4_dcl_temps(tpf, temp_count);
LIST_FOR_EACH_ENTRY(scope, &ctx->scopes, struct hlsl_scope, entry)
{
LIST_FOR_EACH_ENTRY(var, &scope->vars, struct hlsl_ir_var, scope_entry)
{
if (var->is_uniform || var->is_input_semantic || var->is_output_semantic)
continue;
if (!var->regs[HLSL_REGSET_NUMERIC].allocated)
continue;
if (var->indexable)
{
unsigned int id = var->regs[HLSL_REGSET_NUMERIC].id;
unsigned int size = align(var->data_type->reg_size[HLSL_REGSET_NUMERIC], 4) / 4;
write_sm4_dcl_indexable_temp(tpf, id, size, 4);
}
}
}
write_sm4_block(tpf, &func->body);
write_sm4_ret(tpf);
}
static void tpf_write_shdr(struct tpf_compiler *tpf, struct hlsl_ir_function_decl *entry_func) static void tpf_write_shdr(struct tpf_compiler *tpf, struct hlsl_ir_function_decl *entry_func)
{ {
const struct vkd3d_shader_version *version = &tpf->program->shader_version; const struct vkd3d_shader_version *version = &tpf->program->shader_version;
@ -6504,10 +6554,7 @@ static void tpf_write_shdr(struct tpf_compiler *tpf, struct hlsl_ir_function_dec
unsigned int extern_resources_count, i; unsigned int extern_resources_count, i;
const struct hlsl_buffer *cbuffer; const struct hlsl_buffer *cbuffer;
struct hlsl_ctx *ctx = tpf->ctx; struct hlsl_ctx *ctx = tpf->ctx;
const struct hlsl_scope *scope;
const struct hlsl_ir_var *var;
size_t token_count_position; size_t token_count_position;
uint32_t temp_count;
static const uint16_t shader_types[VKD3D_SHADER_TYPE_COUNT] = static const uint16_t shader_types[VKD3D_SHADER_TYPE_COUNT] =
{ {
@ -6522,11 +6569,6 @@ static void tpf_write_shdr(struct tpf_compiler *tpf, struct hlsl_ir_function_dec
VKD3D_SM4_LIB, VKD3D_SM4_LIB,
}; };
mark_indexable_vars(ctx, entry_func);
temp_count = allocate_temp_registers(ctx, entry_func);
if (ctx->result)
return;
tpf->buffer = &buffer; tpf->buffer = &buffer;
extern_resources = sm4_get_extern_resources(ctx, &extern_resources_count); extern_resources = sm4_get_extern_resources(ctx, &extern_resources_count);
@ -6569,40 +6611,7 @@ static void tpf_write_shdr(struct tpf_compiler *tpf, struct hlsl_ir_function_dec
if (version->type == VKD3D_SHADER_TYPE_HULL) if (version->type == VKD3D_SHADER_TYPE_HULL)
tpf_write_hs_control_point_phase(tpf); tpf_write_hs_control_point_phase(tpf);
LIST_FOR_EACH_ENTRY(var, &entry_func->extern_vars, struct hlsl_ir_var, extern_entry) tpf_write_shader_function(tpf, entry_func);
{
if ((var->is_input_semantic && var->last_read) || (var->is_output_semantic && var->first_write))
write_sm4_dcl_semantic(tpf, var);
}
if (version->type == VKD3D_SHADER_TYPE_COMPUTE)
write_sm4_dcl_thread_group(tpf, ctx->thread_count);
if (temp_count)
write_sm4_dcl_temps(tpf, temp_count);
LIST_FOR_EACH_ENTRY(scope, &ctx->scopes, struct hlsl_scope, entry)
{
LIST_FOR_EACH_ENTRY(var, &scope->vars, struct hlsl_ir_var, scope_entry)
{
if (var->is_uniform || var->is_input_semantic || var->is_output_semantic)
continue;
if (!var->regs[HLSL_REGSET_NUMERIC].allocated)
continue;
if (var->indexable)
{
unsigned int id = var->regs[HLSL_REGSET_NUMERIC].id;
unsigned int size = align(var->data_type->reg_size[HLSL_REGSET_NUMERIC], 4) / 4;
write_sm4_dcl_indexable_temp(tpf, id, size, 4);
}
}
}
write_sm4_block(tpf, &entry_func->body);
write_sm4_ret(tpf);
set_u32(&buffer, token_count_position, bytecode_get_size(&buffer) / sizeof(uint32_t)); set_u32(&buffer, token_count_position, bytecode_get_size(&buffer) / sizeof(uint32_t));