KonstantinSeurer
/
mesa
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

aco/isel: Move add_startpgm to aco_instruction_selection.cpp 

Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6504>
This commit is contained in:
Tony Wasserka 2020-09-03 14:28:56 +02:00 committed by Marge Bot
parent 47de553283
commit 793dc668ea
2 changed files with 59 additions and 59 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

59
src/amd/compiler/aco_instruction_selection.cpp
View File

@ -10391,6 +10391,65 @@ static void emit_streamout(isel_context *ctx, unsigned stream)
end_divergent_if(ctx, &ic);
}
Pseudo_instruction *add_startpgm(struct isel_context *ctx)
{
unsigned arg_count = ctx->args->ac.arg_count;
if (ctx->stage == fragment_fs) {
/* LLVM optimizes away unused FS inputs and computes spi_ps_input_addr
* itself and then communicates the results back via the ELF binary.
* Mirror what LLVM does by re-mapping the VGPR arguments here.
*
* TODO: If we made the FS input scanning code into a separate pass that
* could run before argument setup, then this wouldn't be necessary
* anymore.
*/
struct ac_shader_args *args = &ctx->args->ac;
arg_count = 0;
for (unsigned i = 0, vgpr_arg = 0, vgpr_reg = 0; i < args->arg_count; i++) {
if (args->args[i].file != AC_ARG_VGPR) {
arg_count++;
continue;
}
if (!(ctx->program->config->spi_ps_input_addr & (1 << vgpr_arg))) {
args->args[i].skip = true;
} else {
args->args[i].offset = vgpr_reg;
vgpr_reg += args->args[i].size;
arg_count++;
}
vgpr_arg++;
}
}
aco_ptr<Pseudo_instruction> startpgm{create_instruction<Pseudo_instruction>(aco_opcode::p_startpgm, Format::PSEUDO, 0, arg_count + 1)};
for (unsigned i = 0, arg = 0; i < ctx->args->ac.arg_count; i++) {
if (ctx->args->ac.args[i].skip)
continue;
enum ac_arg_regfile file = ctx->args->ac.args[i].file;
unsigned size = ctx->args->ac.args[i].size;
unsigned reg = ctx->args->ac.args[i].offset;
RegClass type = RegClass(file == AC_ARG_SGPR ? RegType::sgpr : RegType::vgpr, size);
Temp dst = Temp{ctx->program->allocateId(), type};
ctx->arg_temps[i] = dst;
startpgm->definitions[arg] = Definition(dst);
startpgm->definitions[arg].setFixed(PhysReg{file == AC_ARG_SGPR ? reg : reg + 256});
arg++;
}
startpgm->definitions[arg_count] = Definition{ctx->program->allocateId(), exec, ctx->program->lane_mask};
Pseudo_instruction *instr = startpgm.get();
ctx->block->instructions.push_back(std::move(startpgm));
/* Stash these in the program so that they can be accessed later when
* handling spilling.
*/
ctx->program->private_segment_buffer = get_arg(ctx, ctx->args->ring_offsets);
ctx->program->scratch_offset = get_arg(ctx, ctx->args->scratch_offset);
return instr;
}
} /* end namespace */
void fix_ls_vgpr_init_bug(isel_context *ctx, Pseudo_instruction *startpgm)

59
src/amd/compiler/aco_instruction_selection.h
View File

@ -201,65 +201,6 @@ inline bool can_subdword_ssbo_store_use_smem(nir_intrinsic_instr *intrin)
void init_context(isel_context *ctx, nir_shader *shader);
inline Pseudo_instruction *add_startpgm(struct isel_context *ctx)
{
unsigned arg_count = ctx->args->ac.arg_count;
if (ctx->stage == fragment_fs) {
/* LLVM optimizes away unused FS inputs and computes spi_ps_input_addr
* itself and then communicates the results back via the ELF binary.
* Mirror what LLVM does by re-mapping the VGPR arguments here.
*
* TODO: If we made the FS input scanning code into a separate pass that
* could run before argument setup, then this wouldn't be necessary
* anymore.
*/
struct ac_shader_args *args = &ctx->args->ac;
arg_count = 0;
for (unsigned i = 0, vgpr_arg = 0, vgpr_reg = 0; i < args->arg_count; i++) {
if (args->args[i].file != AC_ARG_VGPR) {
arg_count++;
continue;
}
if (!(ctx->program->config->spi_ps_input_addr & (1 << vgpr_arg))) {
args->args[i].skip = true;
} else {
args->args[i].offset = vgpr_reg;
vgpr_reg += args->args[i].size;
arg_count++;
}
vgpr_arg++;
}
}
aco_ptr<Pseudo_instruction> startpgm{create_instruction<Pseudo_instruction>(aco_opcode::p_startpgm, Format::PSEUDO, 0, arg_count + 1)};
for (unsigned i = 0, arg = 0; i < ctx->args->ac.arg_count; i++) {
if (ctx->args->ac.args[i].skip)
continue;
enum ac_arg_regfile file = ctx->args->ac.args[i].file;
unsigned size = ctx->args->ac.args[i].size;
unsigned reg = ctx->args->ac.args[i].offset;
RegClass type = RegClass(file == AC_ARG_SGPR ? RegType::sgpr : RegType::vgpr, size);
Temp dst = Temp{ctx->program->allocateId(), type};
ctx->arg_temps[i] = dst;
startpgm->definitions[arg] = Definition(dst);
startpgm->definitions[arg].setFixed(PhysReg{file == AC_ARG_SGPR ? reg : reg + 256});
arg++;
}
startpgm->definitions[arg_count] = Definition{ctx->program->allocateId(), exec, ctx->program->lane_mask};
Pseudo_instruction *instr = startpgm.get();
ctx->block->instructions.push_back(std::move(startpgm));
/* Stash these in the program so that they can be accessed later when
* handling spilling.
*/
ctx->program->private_segment_buffer = get_arg(ctx, ctx->args->ring_offsets);
ctx->program->scratch_offset = get_arg(ctx, ctx->args->scratch_offset);
return instr;
}
isel_context
setup_isel_context(Program* program,
unsigned shader_count,