nir: Add multiplier argument to nir_lower_uniforms_to_ubo.

Note that locations can be set in different units, and the multiplier
argument caters to supporting these different units. For example,
st_glsl_to_nir uses dwords (4 bytes) so the multiplier should be 4,
while tgsi_to_nir uses bytes, so the multiplier should be 16.

Signed-Off-By: Timur Kristóf <timur.kristof@gmail.com>
Tested-by: Andre Heider <a.heider@gmail.com>
Tested-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
This commit is contained in:
Timur Kristóf 2019-02-28 10:53:11 +01:00 committed by Eric Anholt
parent 909d1f50f3
commit 6684e039eb
4 changed files with 18 additions and 11 deletions

View File

@ -3036,7 +3036,7 @@ void nir_lower_io_arrays_to_elements_no_indirects(nir_shader *shader,
void nir_lower_io_to_scalar(nir_shader *shader, nir_variable_mode mask);
void nir_lower_io_to_scalar_early(nir_shader *shader, nir_variable_mode mask);
bool nir_lower_uniforms_to_ubo(nir_shader *shader);
bool nir_lower_uniforms_to_ubo(nir_shader *shader, int multiplier);
typedef struct nir_lower_subgroups_options {
uint8_t subgroup_size;

View File

@ -22,18 +22,24 @@
*/
/*
* Remap load_uniform intrinsics to UBO accesses of UBO binding point 0. Both
* the base and the offset are interpreted as 16-byte units.
*
* Remap load_uniform intrinsics to UBO accesses of UBO binding point 0.
* Simultaneously, remap existing UBO accesses by increasing their binding
* point by 1.
*
* Both the base and the offset are interpreted as 16-byte units.
*
* Note that locations can be set in different units, and the multiplier
* argument caters to supporting these different units.
* For example:
* - st_glsl_to_nir uses dwords (4 bytes) so the multiplier should be 4
* - tgsi_to_nir uses bytes, so the multiplier should be 16
*/
#include "nir.h"
#include "nir_builder.h"
static bool
lower_instr(nir_intrinsic_instr *instr, nir_builder *b)
lower_instr(nir_intrinsic_instr *instr, nir_builder *b, int multiplier)
{
b->cursor = nir_before_instr(&instr->instr);
@ -48,8 +54,8 @@ lower_instr(nir_intrinsic_instr *instr, nir_builder *b)
if (instr->intrinsic == nir_intrinsic_load_uniform) {
nir_ssa_def *ubo_idx = nir_imm_int(b, 0);
nir_ssa_def *ubo_offset =
nir_iadd(b, nir_imm_int(b, 4 * nir_intrinsic_base(instr)),
nir_imul(b, nir_imm_int(b, 4),
nir_iadd(b, nir_imm_int(b, multiplier * nir_intrinsic_base(instr)),
nir_imul(b, nir_imm_int(b, multiplier),
nir_ssa_for_src(b, instr->src[0], 1)));
nir_intrinsic_instr *load =
@ -71,7 +77,7 @@ lower_instr(nir_intrinsic_instr *instr, nir_builder *b)
}
bool
nir_lower_uniforms_to_ubo(nir_shader *shader)
nir_lower_uniforms_to_ubo(nir_shader *shader, int multiplier)
{
bool progress = false;
@ -83,7 +89,8 @@ nir_lower_uniforms_to_ubo(nir_shader *shader)
nir_foreach_instr_safe(instr, block) {
if (instr->type == nir_instr_type_intrinsic)
progress |= lower_instr(nir_instr_as_intrinsic(instr),
&builder);
&builder,
multiplier);
}
}

View File

@ -991,7 +991,7 @@ st_finalize_nir(struct st_context *st, struct gl_program *prog,
if (st->ctx->Const.PackedDriverUniformStorage) {
NIR_PASS_V(nir, nir_lower_io, nir_var_uniform, st_glsl_type_dword_size,
(nir_lower_io_options)0);
NIR_PASS_V(nir, nir_lower_uniforms_to_ubo);
NIR_PASS_V(nir, nir_lower_uniforms_to_ubo, 4);
}
st_nir_lower_samplers(screen, nir, shader_program, prog);

View File

@ -66,7 +66,7 @@ st_nir_finish_builtin_shader(struct st_context *st,
if (st->ctx->Const.PackedDriverUniformStorage) {
NIR_PASS_V(nir, nir_lower_io, nir_var_uniform, st_glsl_type_dword_size,
(nir_lower_io_options)0);
NIR_PASS_V(nir, nir_lower_uniforms_to_ubo);
NIR_PASS_V(nir, nir_lower_uniforms_to_ubo, 4);
}
struct pipe_shader_state state = {