mesa/src/compiler/nir/nir_lower_io.c

566 lines
18 KiB
C
Raw Normal View History

/*
* Copyright © 2014 Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
* IN THE SOFTWARE.
*
* Authors:
* Connor Abbott (cwabbott0@gmail.com)
* Jason Ekstrand (jason@jlekstrand.net)
*
*/
/*
* This lowering pass converts references to input/output variables with
* loads/stores to actual input/output intrinsics.
*/
#include "nir.h"
#include "nir_builder.h"
struct lower_io_state {
nir_builder builder;
int (*type_size)(const struct glsl_type *type);
nir_variable_mode modes;
nir_lower_io_options options;
};
void
nir_assign_var_locations(struct exec_list *var_list, unsigned *size,
int (*type_size)(const struct glsl_type *))
{
unsigned location = 0;
nir_foreach_variable(var, var_list) {
/*
* UBOs have their own address spaces, so don't count them towards the
* number of global uniforms
*/
if ((var->data.mode == nir_var_uniform || var->data.mode == nir_var_shader_storage) &&
var->interface_type != NULL)
continue;
var->data.driver_location = location;
location += type_size(var->type);
}
*size = location;
}
/**
* Return true if the given variable is a per-vertex input/output array.
* (such as geometry shader inputs).
*/
bool
nir_is_per_vertex_io(nir_variable *var, gl_shader_stage stage)
{
if (var->data.patch || !glsl_type_is_array(var->type))
return false;
if (var->data.mode == nir_var_shader_in)
return stage == MESA_SHADER_GEOMETRY ||
stage == MESA_SHADER_TESS_CTRL ||
stage == MESA_SHADER_TESS_EVAL;
if (var->data.mode == nir_var_shader_out)
return stage == MESA_SHADER_TESS_CTRL;
return false;
}
nir: Get rid of *_indirect variants of input/output load/store intrinsics There is some special-casing needed in a competent back-end. However, they can do their special-casing easily enough based on whether or not the offset is a constant. In the mean time, having the *_indirect variants adds special cases a number of places where they don't need to be and, in general, only complicates things. To complicate matters, NIR had no way to convdert an indirect load/store to a direct one in the case that the indirect was a constant so we would still not really get what the back-ends wanted. The best solution seems to be to get rid of the *_indirect variants entirely. This commit is a bunch of different changes squashed together: - nir: Get rid of *_indirect variants of input/output load/store intrinsics - nir/glsl: Stop handling UBO/SSBO load/stores differently depending on indirect - nir/lower_io: Get rid of load/store_foo_indirect - i965/fs: Get rid of load/store_foo_indirect - i965/vec4: Get rid of load/store_foo_indirect - tgsi_to_nir: Get rid of load/store_foo_indirect - ir3/nir: Use the new unified io intrinsics - vc4: Do all uniform loads with byte offsets - vc4/nir: Use the new unified io intrinsics - vc4: Fix load_user_clip_plane crash - vc4: add missing src for store outputs - vc4: Fix state uniforms - nir/lower_clip: Update to the new load/store intrinsics - nir/lower_two_sided_color: Update to the new load intrinsic NIR and i965 changes are Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> NIR indirect declarations and vc4 changes are Reviewed-by: Eric Anholt <eric@anholt.net> ir3 changes are Reviewed-by: Rob Clark <robdclark@gmail.com> NIR changes are Acked-by: Rob Clark <robdclark@gmail.com>
2015-11-25 22:14:05 +00:00
static nir_ssa_def *
get_io_offset(nir_builder *b, nir_deref_var *deref,
nir_ssa_def **vertex_index,
int (*type_size)(const struct glsl_type *),
unsigned *component)
{
nir_deref *tail = &deref->deref;
/* For per-vertex input arrays (i.e. geometry shader inputs), keep the
* outermost array index separate. Process the rest normally.
*/
if (vertex_index != NULL) {
tail = tail->child;
nir_deref_array *deref_array = nir_deref_as_array(tail);
nir_ssa_def *vtx = nir_imm_int(b, deref_array->base_offset);
if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
vtx = nir_iadd(b, vtx, nir_ssa_for_src(b, deref_array->indirect, 1));
}
*vertex_index = vtx;
}
if (deref->var->data.compact) {
assert(tail->child->deref_type == nir_deref_type_array);
assert(glsl_type_is_scalar(glsl_without_array(deref->var->type)));
nir_deref_array *deref_array = nir_deref_as_array(tail->child);
/* We always lower indirect dereferences for "compact" array vars. */
assert(deref_array->deref_array_type == nir_deref_array_type_direct);
const unsigned total_offset = *component + deref_array->base_offset;
const unsigned slot_offset = total_offset / 4;
*component = total_offset % 4;
return nir_imm_int(b, type_size(glsl_vec4_type()) * slot_offset);
}
nir: Get rid of *_indirect variants of input/output load/store intrinsics There is some special-casing needed in a competent back-end. However, they can do their special-casing easily enough based on whether or not the offset is a constant. In the mean time, having the *_indirect variants adds special cases a number of places where they don't need to be and, in general, only complicates things. To complicate matters, NIR had no way to convdert an indirect load/store to a direct one in the case that the indirect was a constant so we would still not really get what the back-ends wanted. The best solution seems to be to get rid of the *_indirect variants entirely. This commit is a bunch of different changes squashed together: - nir: Get rid of *_indirect variants of input/output load/store intrinsics - nir/glsl: Stop handling UBO/SSBO load/stores differently depending on indirect - nir/lower_io: Get rid of load/store_foo_indirect - i965/fs: Get rid of load/store_foo_indirect - i965/vec4: Get rid of load/store_foo_indirect - tgsi_to_nir: Get rid of load/store_foo_indirect - ir3/nir: Use the new unified io intrinsics - vc4: Do all uniform loads with byte offsets - vc4/nir: Use the new unified io intrinsics - vc4: Fix load_user_clip_plane crash - vc4: add missing src for store outputs - vc4: Fix state uniforms - nir/lower_clip: Update to the new load/store intrinsics - nir/lower_two_sided_color: Update to the new load intrinsic NIR and i965 changes are Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> NIR indirect declarations and vc4 changes are Reviewed-by: Eric Anholt <eric@anholt.net> ir3 changes are Reviewed-by: Rob Clark <robdclark@gmail.com> NIR changes are Acked-by: Rob Clark <robdclark@gmail.com>
2015-11-25 22:14:05 +00:00
/* Just emit code and let constant-folding go to town */
nir_ssa_def *offset = nir_imm_int(b, 0);
while (tail->child != NULL) {
const struct glsl_type *parent_type = tail->type;
tail = tail->child;
if (tail->deref_type == nir_deref_type_array) {
nir_deref_array *deref_array = nir_deref_as_array(tail);
unsigned size = type_size(tail->type);
nir: Get rid of *_indirect variants of input/output load/store intrinsics There is some special-casing needed in a competent back-end. However, they can do their special-casing easily enough based on whether or not the offset is a constant. In the mean time, having the *_indirect variants adds special cases a number of places where they don't need to be and, in general, only complicates things. To complicate matters, NIR had no way to convdert an indirect load/store to a direct one in the case that the indirect was a constant so we would still not really get what the back-ends wanted. The best solution seems to be to get rid of the *_indirect variants entirely. This commit is a bunch of different changes squashed together: - nir: Get rid of *_indirect variants of input/output load/store intrinsics - nir/glsl: Stop handling UBO/SSBO load/stores differently depending on indirect - nir/lower_io: Get rid of load/store_foo_indirect - i965/fs: Get rid of load/store_foo_indirect - i965/vec4: Get rid of load/store_foo_indirect - tgsi_to_nir: Get rid of load/store_foo_indirect - ir3/nir: Use the new unified io intrinsics - vc4: Do all uniform loads with byte offsets - vc4/nir: Use the new unified io intrinsics - vc4: Fix load_user_clip_plane crash - vc4: add missing src for store outputs - vc4: Fix state uniforms - nir/lower_clip: Update to the new load/store intrinsics - nir/lower_two_sided_color: Update to the new load intrinsic NIR and i965 changes are Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> NIR indirect declarations and vc4 changes are Reviewed-by: Eric Anholt <eric@anholt.net> ir3 changes are Reviewed-by: Rob Clark <robdclark@gmail.com> NIR changes are Acked-by: Rob Clark <robdclark@gmail.com>
2015-11-25 22:14:05 +00:00
offset = nir_iadd(b, offset,
nir_imm_int(b, size * deref_array->base_offset));
if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
nir_ssa_def *mul =
nir_imul(b, nir_imm_int(b, size),
nir_ssa_for_src(b, deref_array->indirect, 1));
nir: Get rid of *_indirect variants of input/output load/store intrinsics There is some special-casing needed in a competent back-end. However, they can do their special-casing easily enough based on whether or not the offset is a constant. In the mean time, having the *_indirect variants adds special cases a number of places where they don't need to be and, in general, only complicates things. To complicate matters, NIR had no way to convdert an indirect load/store to a direct one in the case that the indirect was a constant so we would still not really get what the back-ends wanted. The best solution seems to be to get rid of the *_indirect variants entirely. This commit is a bunch of different changes squashed together: - nir: Get rid of *_indirect variants of input/output load/store intrinsics - nir/glsl: Stop handling UBO/SSBO load/stores differently depending on indirect - nir/lower_io: Get rid of load/store_foo_indirect - i965/fs: Get rid of load/store_foo_indirect - i965/vec4: Get rid of load/store_foo_indirect - tgsi_to_nir: Get rid of load/store_foo_indirect - ir3/nir: Use the new unified io intrinsics - vc4: Do all uniform loads with byte offsets - vc4/nir: Use the new unified io intrinsics - vc4: Fix load_user_clip_plane crash - vc4: add missing src for store outputs - vc4: Fix state uniforms - nir/lower_clip: Update to the new load/store intrinsics - nir/lower_two_sided_color: Update to the new load intrinsic NIR and i965 changes are Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> NIR indirect declarations and vc4 changes are Reviewed-by: Eric Anholt <eric@anholt.net> ir3 changes are Reviewed-by: Rob Clark <robdclark@gmail.com> NIR changes are Acked-by: Rob Clark <robdclark@gmail.com>
2015-11-25 22:14:05 +00:00
offset = nir_iadd(b, offset, mul);
}
} else if (tail->deref_type == nir_deref_type_struct) {
nir_deref_struct *deref_struct = nir_deref_as_struct(tail);
nir: Get rid of *_indirect variants of input/output load/store intrinsics There is some special-casing needed in a competent back-end. However, they can do their special-casing easily enough based on whether or not the offset is a constant. In the mean time, having the *_indirect variants adds special cases a number of places where they don't need to be and, in general, only complicates things. To complicate matters, NIR had no way to convdert an indirect load/store to a direct one in the case that the indirect was a constant so we would still not really get what the back-ends wanted. The best solution seems to be to get rid of the *_indirect variants entirely. This commit is a bunch of different changes squashed together: - nir: Get rid of *_indirect variants of input/output load/store intrinsics - nir/glsl: Stop handling UBO/SSBO load/stores differently depending on indirect - nir/lower_io: Get rid of load/store_foo_indirect - i965/fs: Get rid of load/store_foo_indirect - i965/vec4: Get rid of load/store_foo_indirect - tgsi_to_nir: Get rid of load/store_foo_indirect - ir3/nir: Use the new unified io intrinsics - vc4: Do all uniform loads with byte offsets - vc4/nir: Use the new unified io intrinsics - vc4: Fix load_user_clip_plane crash - vc4: add missing src for store outputs - vc4: Fix state uniforms - nir/lower_clip: Update to the new load/store intrinsics - nir/lower_two_sided_color: Update to the new load intrinsic NIR and i965 changes are Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> NIR indirect declarations and vc4 changes are Reviewed-by: Eric Anholt <eric@anholt.net> ir3 changes are Reviewed-by: Rob Clark <robdclark@gmail.com> NIR changes are Acked-by: Rob Clark <robdclark@gmail.com>
2015-11-25 22:14:05 +00:00
unsigned field_offset = 0;
for (unsigned i = 0; i < deref_struct->index; i++) {
nir: Get rid of *_indirect variants of input/output load/store intrinsics There is some special-casing needed in a competent back-end. However, they can do their special-casing easily enough based on whether or not the offset is a constant. In the mean time, having the *_indirect variants adds special cases a number of places where they don't need to be and, in general, only complicates things. To complicate matters, NIR had no way to convdert an indirect load/store to a direct one in the case that the indirect was a constant so we would still not really get what the back-ends wanted. The best solution seems to be to get rid of the *_indirect variants entirely. This commit is a bunch of different changes squashed together: - nir: Get rid of *_indirect variants of input/output load/store intrinsics - nir/glsl: Stop handling UBO/SSBO load/stores differently depending on indirect - nir/lower_io: Get rid of load/store_foo_indirect - i965/fs: Get rid of load/store_foo_indirect - i965/vec4: Get rid of load/store_foo_indirect - tgsi_to_nir: Get rid of load/store_foo_indirect - ir3/nir: Use the new unified io intrinsics - vc4: Do all uniform loads with byte offsets - vc4/nir: Use the new unified io intrinsics - vc4: Fix load_user_clip_plane crash - vc4: add missing src for store outputs - vc4: Fix state uniforms - nir/lower_clip: Update to the new load/store intrinsics - nir/lower_two_sided_color: Update to the new load intrinsic NIR and i965 changes are Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> NIR indirect declarations and vc4 changes are Reviewed-by: Eric Anholt <eric@anholt.net> ir3 changes are Reviewed-by: Rob Clark <robdclark@gmail.com> NIR changes are Acked-by: Rob Clark <robdclark@gmail.com>
2015-11-25 22:14:05 +00:00
field_offset += type_size(glsl_get_struct_field(parent_type, i));
}
nir: Get rid of *_indirect variants of input/output load/store intrinsics There is some special-casing needed in a competent back-end. However, they can do their special-casing easily enough based on whether or not the offset is a constant. In the mean time, having the *_indirect variants adds special cases a number of places where they don't need to be and, in general, only complicates things. To complicate matters, NIR had no way to convdert an indirect load/store to a direct one in the case that the indirect was a constant so we would still not really get what the back-ends wanted. The best solution seems to be to get rid of the *_indirect variants entirely. This commit is a bunch of different changes squashed together: - nir: Get rid of *_indirect variants of input/output load/store intrinsics - nir/glsl: Stop handling UBO/SSBO load/stores differently depending on indirect - nir/lower_io: Get rid of load/store_foo_indirect - i965/fs: Get rid of load/store_foo_indirect - i965/vec4: Get rid of load/store_foo_indirect - tgsi_to_nir: Get rid of load/store_foo_indirect - ir3/nir: Use the new unified io intrinsics - vc4: Do all uniform loads with byte offsets - vc4/nir: Use the new unified io intrinsics - vc4: Fix load_user_clip_plane crash - vc4: add missing src for store outputs - vc4: Fix state uniforms - nir/lower_clip: Update to the new load/store intrinsics - nir/lower_two_sided_color: Update to the new load intrinsic NIR and i965 changes are Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> NIR indirect declarations and vc4 changes are Reviewed-by: Eric Anholt <eric@anholt.net> ir3 changes are Reviewed-by: Rob Clark <robdclark@gmail.com> NIR changes are Acked-by: Rob Clark <robdclark@gmail.com>
2015-11-25 22:14:05 +00:00
offset = nir_iadd(b, offset, nir_imm_int(b, field_offset));
}
}
nir: Get rid of *_indirect variants of input/output load/store intrinsics There is some special-casing needed in a competent back-end. However, they can do their special-casing easily enough based on whether or not the offset is a constant. In the mean time, having the *_indirect variants adds special cases a number of places where they don't need to be and, in general, only complicates things. To complicate matters, NIR had no way to convdert an indirect load/store to a direct one in the case that the indirect was a constant so we would still not really get what the back-ends wanted. The best solution seems to be to get rid of the *_indirect variants entirely. This commit is a bunch of different changes squashed together: - nir: Get rid of *_indirect variants of input/output load/store intrinsics - nir/glsl: Stop handling UBO/SSBO load/stores differently depending on indirect - nir/lower_io: Get rid of load/store_foo_indirect - i965/fs: Get rid of load/store_foo_indirect - i965/vec4: Get rid of load/store_foo_indirect - tgsi_to_nir: Get rid of load/store_foo_indirect - ir3/nir: Use the new unified io intrinsics - vc4: Do all uniform loads with byte offsets - vc4/nir: Use the new unified io intrinsics - vc4: Fix load_user_clip_plane crash - vc4: add missing src for store outputs - vc4: Fix state uniforms - nir/lower_clip: Update to the new load/store intrinsics - nir/lower_two_sided_color: Update to the new load intrinsic NIR and i965 changes are Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> NIR indirect declarations and vc4 changes are Reviewed-by: Eric Anholt <eric@anholt.net> ir3 changes are Reviewed-by: Rob Clark <robdclark@gmail.com> NIR changes are Acked-by: Rob Clark <robdclark@gmail.com>
2015-11-25 22:14:05 +00:00
return offset;
}
static nir_intrinsic_instr *
lower_load(nir_intrinsic_instr *intrin, struct lower_io_state *state,
nir_ssa_def *vertex_index, nir_ssa_def *offset,
unsigned component)
{
const nir_shader *nir = state->builder.shader;
nir_variable *var = intrin->variables[0]->var;
nir_variable_mode mode = var->data.mode;
nir_ssa_def *barycentric = NULL;
nir_intrinsic_op op;
switch (mode) {
case nir_var_shader_in:
if (nir->stage == MESA_SHADER_FRAGMENT &&
nir->options->use_interpolated_input_intrinsics &&
var->data.interpolation != INTERP_MODE_FLAT) {
assert(vertex_index == NULL);
nir_intrinsic_op bary_op;
if (var->data.sample ||
(state->options & nir_lower_io_force_sample_interpolation))
bary_op = nir_intrinsic_load_barycentric_sample;
else if (var->data.centroid)
bary_op = nir_intrinsic_load_barycentric_centroid;
else
bary_op = nir_intrinsic_load_barycentric_pixel;
barycentric = nir_load_barycentric(&state->builder, bary_op,
var->data.interpolation);
op = nir_intrinsic_load_interpolated_input;
} else {
op = vertex_index ? nir_intrinsic_load_per_vertex_input :
nir_intrinsic_load_input;
}
break;
case nir_var_shader_out:
op = vertex_index ? nir_intrinsic_load_per_vertex_output :
nir_intrinsic_load_output;
break;
case nir_var_uniform:
nir: Get rid of *_indirect variants of input/output load/store intrinsics There is some special-casing needed in a competent back-end. However, they can do their special-casing easily enough based on whether or not the offset is a constant. In the mean time, having the *_indirect variants adds special cases a number of places where they don't need to be and, in general, only complicates things. To complicate matters, NIR had no way to convdert an indirect load/store to a direct one in the case that the indirect was a constant so we would still not really get what the back-ends wanted. The best solution seems to be to get rid of the *_indirect variants entirely. This commit is a bunch of different changes squashed together: - nir: Get rid of *_indirect variants of input/output load/store intrinsics - nir/glsl: Stop handling UBO/SSBO load/stores differently depending on indirect - nir/lower_io: Get rid of load/store_foo_indirect - i965/fs: Get rid of load/store_foo_indirect - i965/vec4: Get rid of load/store_foo_indirect - tgsi_to_nir: Get rid of load/store_foo_indirect - ir3/nir: Use the new unified io intrinsics - vc4: Do all uniform loads with byte offsets - vc4/nir: Use the new unified io intrinsics - vc4: Fix load_user_clip_plane crash - vc4: add missing src for store outputs - vc4: Fix state uniforms - nir/lower_clip: Update to the new load/store intrinsics - nir/lower_two_sided_color: Update to the new load intrinsic NIR and i965 changes are Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> NIR indirect declarations and vc4 changes are Reviewed-by: Eric Anholt <eric@anholt.net> ir3 changes are Reviewed-by: Rob Clark <robdclark@gmail.com> NIR changes are Acked-by: Rob Clark <robdclark@gmail.com>
2015-11-25 22:14:05 +00:00
op = nir_intrinsic_load_uniform;
break;
case nir_var_shared:
op = nir_intrinsic_load_shared;
break;
default:
unreachable("Unknown variable mode");
}
nir_intrinsic_instr *load =
nir_intrinsic_instr_create(state->builder.shader, op);
load->num_components = intrin->num_components;
nir_intrinsic_set_base(load, var->data.driver_location);
if (mode == nir_var_shader_in || mode == nir_var_shader_out)
nir_intrinsic_set_component(load, component);
if (load->intrinsic == nir_intrinsic_load_uniform)
nir_intrinsic_set_range(load, state->type_size(var->type));
if (vertex_index) {
load->src[0] = nir_src_for_ssa(vertex_index);
load->src[1] = nir_src_for_ssa(offset);
} else if (barycentric) {
load->src[0] = nir_src_for_ssa(barycentric);
load->src[1] = nir_src_for_ssa(offset);
} else {
load->src[0] = nir_src_for_ssa(offset);
}
return load;
}
static nir_intrinsic_instr *
lower_store(nir_intrinsic_instr *intrin, struct lower_io_state *state,
nir_ssa_def *vertex_index, nir_ssa_def *offset,
unsigned component)
{
nir_variable *var = intrin->variables[0]->var;
nir_variable_mode mode = var->data.mode;
nir_intrinsic_op op;
if (mode == nir_var_shared) {
op = nir_intrinsic_store_shared;
} else {
assert(mode == nir_var_shader_out);
op = vertex_index ? nir_intrinsic_store_per_vertex_output :
nir_intrinsic_store_output;
}
nir_intrinsic_instr *store =
nir_intrinsic_instr_create(state->builder.shader, op);
store->num_components = intrin->num_components;
nir_src_copy(&store->src[0], &intrin->src[0], store);
nir_intrinsic_set_base(store, var->data.driver_location);
if (mode == nir_var_shader_out)
nir_intrinsic_set_component(store, component);
nir_intrinsic_set_write_mask(store, nir_intrinsic_write_mask(intrin));
if (vertex_index)
store->src[1] = nir_src_for_ssa(vertex_index);
store->src[vertex_index ? 2 : 1] = nir_src_for_ssa(offset);
return store;
}
static nir_intrinsic_instr *
lower_atomic(nir_intrinsic_instr *intrin, struct lower_io_state *state,
nir_ssa_def *offset)
{
nir_variable *var = intrin->variables[0]->var;
assert(var->data.mode == nir_var_shared);
nir_intrinsic_op op;
switch (intrin->intrinsic) {
#define OP(O) case nir_intrinsic_var_##O: op = nir_intrinsic_shared_##O; break;
OP(atomic_exchange)
OP(atomic_comp_swap)
OP(atomic_add)
OP(atomic_imin)
OP(atomic_umin)
OP(atomic_imax)
OP(atomic_umax)
OP(atomic_and)
OP(atomic_or)
OP(atomic_xor)
#undef OP
default:
unreachable("Invalid atomic");
}
nir_intrinsic_instr *atomic =
nir_intrinsic_instr_create(state->builder.shader, op);
nir_intrinsic_set_base(atomic, var->data.driver_location);
atomic->src[0] = nir_src_for_ssa(offset);
for (unsigned i = 0; i < nir_intrinsic_infos[intrin->intrinsic].num_srcs; i++) {
nir_src_copy(&atomic->src[i+1], &intrin->src[i], atomic);
}
return atomic;
}
static nir_intrinsic_instr *
lower_interpolate_at(nir_intrinsic_instr *intrin, struct lower_io_state *state,
nir_ssa_def *offset, unsigned component)
{
nir_variable *var = intrin->variables[0]->var;
assert(var->data.mode == nir_var_shader_in);
/* Ignore interpolateAt() for flat variables - flat is flat. */
if (var->data.interpolation == INTERP_MODE_FLAT)
return lower_load(intrin, state, NULL, offset, component);
nir_intrinsic_op bary_op;
switch (intrin->intrinsic) {
case nir_intrinsic_interp_var_at_centroid:
bary_op = (state->options & nir_lower_io_force_sample_interpolation) ?
nir_intrinsic_load_barycentric_sample :
nir_intrinsic_load_barycentric_centroid;
break;
case nir_intrinsic_interp_var_at_sample:
bary_op = nir_intrinsic_load_barycentric_at_sample;
break;
case nir_intrinsic_interp_var_at_offset:
bary_op = nir_intrinsic_load_barycentric_at_offset;
break;
default:
unreachable("Bogus interpolateAt() intrinsic.");
}
nir_intrinsic_instr *bary_setup =
nir_intrinsic_instr_create(state->builder.shader, bary_op);
nir_ssa_dest_init(&bary_setup->instr, &bary_setup->dest, 2, 32, NULL);
nir_intrinsic_set_interp_mode(bary_setup, var->data.interpolation);
if (intrin->intrinsic != nir_intrinsic_interp_var_at_centroid)
nir_src_copy(&bary_setup->src[0], &intrin->src[0], bary_setup);
nir_builder_instr_insert(&state->builder, &bary_setup->instr);
nir_intrinsic_instr *load =
nir_intrinsic_instr_create(state->builder.shader,
nir_intrinsic_load_interpolated_input);
load->num_components = intrin->num_components;
nir_intrinsic_set_base(load, var->data.driver_location);
nir_intrinsic_set_component(load, component);
load->src[0] = nir_src_for_ssa(&bary_setup->dest.ssa);
load->src[1] = nir_src_for_ssa(offset);
return load;
}
static bool
nir_lower_io_block(nir_block *block,
struct lower_io_state *state)
{
nir_builder *b = &state->builder;
const nir_shader_compiler_options *options = b->shader->options;
bool progress = false;
nir_foreach_instr_safe(instr, block) {
if (instr->type != nir_instr_type_intrinsic)
continue;
nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
switch (intrin->intrinsic) {
case nir_intrinsic_load_var:
case nir_intrinsic_store_var:
case nir_intrinsic_var_atomic_add:
case nir_intrinsic_var_atomic_imin:
case nir_intrinsic_var_atomic_umin:
case nir_intrinsic_var_atomic_imax:
case nir_intrinsic_var_atomic_umax:
case nir_intrinsic_var_atomic_and:
case nir_intrinsic_var_atomic_or:
case nir_intrinsic_var_atomic_xor:
case nir_intrinsic_var_atomic_exchange:
case nir_intrinsic_var_atomic_comp_swap:
/* We can lower the io for this nir instrinsic */
break;
case nir_intrinsic_interp_var_at_centroid:
case nir_intrinsic_interp_var_at_sample:
case nir_intrinsic_interp_var_at_offset:
/* We can optionally lower these to load_interpolated_input */
if (options->use_interpolated_input_intrinsics)
break;
default:
/* We can't lower the io for this nir instrinsic, so skip it */
continue;
}
nir_variable *var = intrin->variables[0]->var;
nir_variable_mode mode = var->data.mode;
if ((state->modes & mode) == 0)
continue;
if (mode != nir_var_shader_in &&
mode != nir_var_shader_out &&
mode != nir_var_shared &&
mode != nir_var_uniform)
continue;
b->cursor = nir_before_instr(instr);
const bool per_vertex = nir_is_per_vertex_io(var, b->shader->stage);
nir_ssa_def *offset;
nir_ssa_def *vertex_index = NULL;
unsigned component_offset = var->data.location_frac;
offset = get_io_offset(b, intrin->variables[0],
per_vertex ? &vertex_index : NULL,
state->type_size, &component_offset);
nir_intrinsic_instr *replacement;
switch (intrin->intrinsic) {
case nir_intrinsic_load_var:
replacement = lower_load(intrin, state, vertex_index, offset,
component_offset);
break;
case nir_intrinsic_store_var:
replacement = lower_store(intrin, state, vertex_index, offset,
component_offset);
break;
case nir_intrinsic_var_atomic_add:
case nir_intrinsic_var_atomic_imin:
case nir_intrinsic_var_atomic_umin:
case nir_intrinsic_var_atomic_imax:
case nir_intrinsic_var_atomic_umax:
case nir_intrinsic_var_atomic_and:
case nir_intrinsic_var_atomic_or:
case nir_intrinsic_var_atomic_xor:
case nir_intrinsic_var_atomic_exchange:
case nir_intrinsic_var_atomic_comp_swap:
assert(vertex_index == NULL);
replacement = lower_atomic(intrin, state, offset);
break;
case nir_intrinsic_interp_var_at_centroid:
case nir_intrinsic_interp_var_at_sample:
case nir_intrinsic_interp_var_at_offset:
assert(vertex_index == NULL);
replacement = lower_interpolate_at(intrin, state, offset,
component_offset);
break;
default:
continue;
}
if (nir_intrinsic_infos[intrin->intrinsic].has_dest) {
if (intrin->dest.is_ssa) {
nir_ssa_dest_init(&replacement->instr, &replacement->dest,
intrin->dest.ssa.num_components,
intrin->dest.ssa.bit_size, NULL);
nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
nir_src_for_ssa(&replacement->dest.ssa));
} else {
nir_dest_copy(&replacement->dest, &intrin->dest, &intrin->instr);
}
}
nir_instr_insert_before(&intrin->instr, &replacement->instr);
nir_instr_remove(&intrin->instr);
progress = true;
}
return progress;
}
static bool
nir_lower_io_impl(nir_function_impl *impl,
nir_variable_mode modes,
int (*type_size)(const struct glsl_type *),
nir_lower_io_options options)
{
struct lower_io_state state;
bool progress = false;
nir_builder_init(&state.builder, impl);
state.modes = modes;
state.type_size = type_size;
state.options = options;
nir_foreach_block(block, impl) {
progress |= nir_lower_io_block(block, &state);
}
nir_metadata_preserve(impl, nir_metadata_block_index |
nir_metadata_dominance);
return progress;
}
bool
nir_lower_io(nir_shader *shader, nir_variable_mode modes,
int (*type_size)(const struct glsl_type *),
nir_lower_io_options options)
{
bool progress = false;
nir_foreach_function(function, shader) {
if (function->impl) {
progress |= nir_lower_io_impl(function->impl, modes,
type_size, options);
}
}
return progress;
}
/**
nir: Get rid of *_indirect variants of input/output load/store intrinsics There is some special-casing needed in a competent back-end. However, they can do their special-casing easily enough based on whether or not the offset is a constant. In the mean time, having the *_indirect variants adds special cases a number of places where they don't need to be and, in general, only complicates things. To complicate matters, NIR had no way to convdert an indirect load/store to a direct one in the case that the indirect was a constant so we would still not really get what the back-ends wanted. The best solution seems to be to get rid of the *_indirect variants entirely. This commit is a bunch of different changes squashed together: - nir: Get rid of *_indirect variants of input/output load/store intrinsics - nir/glsl: Stop handling UBO/SSBO load/stores differently depending on indirect - nir/lower_io: Get rid of load/store_foo_indirect - i965/fs: Get rid of load/store_foo_indirect - i965/vec4: Get rid of load/store_foo_indirect - tgsi_to_nir: Get rid of load/store_foo_indirect - ir3/nir: Use the new unified io intrinsics - vc4: Do all uniform loads with byte offsets - vc4/nir: Use the new unified io intrinsics - vc4: Fix load_user_clip_plane crash - vc4: add missing src for store outputs - vc4: Fix state uniforms - nir/lower_clip: Update to the new load/store intrinsics - nir/lower_two_sided_color: Update to the new load intrinsic NIR and i965 changes are Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> NIR indirect declarations and vc4 changes are Reviewed-by: Eric Anholt <eric@anholt.net> ir3 changes are Reviewed-by: Rob Clark <robdclark@gmail.com> NIR changes are Acked-by: Rob Clark <robdclark@gmail.com>
2015-11-25 22:14:05 +00:00
* Return the offset soruce for a load/store intrinsic.
*/
nir_src *
nir: Get rid of *_indirect variants of input/output load/store intrinsics There is some special-casing needed in a competent back-end. However, they can do their special-casing easily enough based on whether or not the offset is a constant. In the mean time, having the *_indirect variants adds special cases a number of places where they don't need to be and, in general, only complicates things. To complicate matters, NIR had no way to convdert an indirect load/store to a direct one in the case that the indirect was a constant so we would still not really get what the back-ends wanted. The best solution seems to be to get rid of the *_indirect variants entirely. This commit is a bunch of different changes squashed together: - nir: Get rid of *_indirect variants of input/output load/store intrinsics - nir/glsl: Stop handling UBO/SSBO load/stores differently depending on indirect - nir/lower_io: Get rid of load/store_foo_indirect - i965/fs: Get rid of load/store_foo_indirect - i965/vec4: Get rid of load/store_foo_indirect - tgsi_to_nir: Get rid of load/store_foo_indirect - ir3/nir: Use the new unified io intrinsics - vc4: Do all uniform loads with byte offsets - vc4/nir: Use the new unified io intrinsics - vc4: Fix load_user_clip_plane crash - vc4: add missing src for store outputs - vc4: Fix state uniforms - nir/lower_clip: Update to the new load/store intrinsics - nir/lower_two_sided_color: Update to the new load intrinsic NIR and i965 changes are Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> NIR indirect declarations and vc4 changes are Reviewed-by: Eric Anholt <eric@anholt.net> ir3 changes are Reviewed-by: Rob Clark <robdclark@gmail.com> NIR changes are Acked-by: Rob Clark <robdclark@gmail.com>
2015-11-25 22:14:05 +00:00
nir_get_io_offset_src(nir_intrinsic_instr *instr)
{
switch (instr->intrinsic) {
nir: Get rid of *_indirect variants of input/output load/store intrinsics There is some special-casing needed in a competent back-end. However, they can do their special-casing easily enough based on whether or not the offset is a constant. In the mean time, having the *_indirect variants adds special cases a number of places where they don't need to be and, in general, only complicates things. To complicate matters, NIR had no way to convdert an indirect load/store to a direct one in the case that the indirect was a constant so we would still not really get what the back-ends wanted. The best solution seems to be to get rid of the *_indirect variants entirely. This commit is a bunch of different changes squashed together: - nir: Get rid of *_indirect variants of input/output load/store intrinsics - nir/glsl: Stop handling UBO/SSBO load/stores differently depending on indirect - nir/lower_io: Get rid of load/store_foo_indirect - i965/fs: Get rid of load/store_foo_indirect - i965/vec4: Get rid of load/store_foo_indirect - tgsi_to_nir: Get rid of load/store_foo_indirect - ir3/nir: Use the new unified io intrinsics - vc4: Do all uniform loads with byte offsets - vc4/nir: Use the new unified io intrinsics - vc4: Fix load_user_clip_plane crash - vc4: add missing src for store outputs - vc4: Fix state uniforms - nir/lower_clip: Update to the new load/store intrinsics - nir/lower_two_sided_color: Update to the new load intrinsic NIR and i965 changes are Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> NIR indirect declarations and vc4 changes are Reviewed-by: Eric Anholt <eric@anholt.net> ir3 changes are Reviewed-by: Rob Clark <robdclark@gmail.com> NIR changes are Acked-by: Rob Clark <robdclark@gmail.com>
2015-11-25 22:14:05 +00:00
case nir_intrinsic_load_input:
case nir_intrinsic_load_output:
case nir_intrinsic_load_uniform:
return &instr->src[0];
case nir_intrinsic_load_ubo:
case nir_intrinsic_load_ssbo:
nir: Get rid of *_indirect variants of input/output load/store intrinsics There is some special-casing needed in a competent back-end. However, they can do their special-casing easily enough based on whether or not the offset is a constant. In the mean time, having the *_indirect variants adds special cases a number of places where they don't need to be and, in general, only complicates things. To complicate matters, NIR had no way to convdert an indirect load/store to a direct one in the case that the indirect was a constant so we would still not really get what the back-ends wanted. The best solution seems to be to get rid of the *_indirect variants entirely. This commit is a bunch of different changes squashed together: - nir: Get rid of *_indirect variants of input/output load/store intrinsics - nir/glsl: Stop handling UBO/SSBO load/stores differently depending on indirect - nir/lower_io: Get rid of load/store_foo_indirect - i965/fs: Get rid of load/store_foo_indirect - i965/vec4: Get rid of load/store_foo_indirect - tgsi_to_nir: Get rid of load/store_foo_indirect - ir3/nir: Use the new unified io intrinsics - vc4: Do all uniform loads with byte offsets - vc4/nir: Use the new unified io intrinsics - vc4: Fix load_user_clip_plane crash - vc4: add missing src for store outputs - vc4: Fix state uniforms - nir/lower_clip: Update to the new load/store intrinsics - nir/lower_two_sided_color: Update to the new load intrinsic NIR and i965 changes are Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> NIR indirect declarations and vc4 changes are Reviewed-by: Eric Anholt <eric@anholt.net> ir3 changes are Reviewed-by: Rob Clark <robdclark@gmail.com> NIR changes are Acked-by: Rob Clark <robdclark@gmail.com>
2015-11-25 22:14:05 +00:00
case nir_intrinsic_load_per_vertex_input:
case nir_intrinsic_load_per_vertex_output:
nir: Add new intrinsics for fragment shader input interpolation. Backends can normally handle shader inputs solely by looking at load_input intrinsics, and ignore the nir_variables in nir->inputs. One exception is fragment shader inputs. load_input doesn't capture the necessary interpolation information - flat, smooth, noperspective mode, and centroid, sample, or pixel for the location. This means that backends have to interpolate based on the nir_variables, then associate those with the load_input intrinsics (say, by storing a map of which variables are at which locations). With GL_ARB_enhanced_layouts, we're going to have multiple varyings packed into a single vec4 location. The intrinsics make this easy: simply load N components from location <loc, component>. However, working with variables and correlating the two is very awkward; we'd much rather have intrinsics capture all the necessary information. Fragment shader input interpolation typically works by producing a set of barycentric coordinates, then using those to do a linear interpolation between the values at the triangle's corners. We represent this by introducing five new load_barycentric_* intrinsics: - load_barycentric_pixel (ordinary variable) - load_barycentric_centroid (centroid qualified variable) - load_barycentric_sample (sample qualified variable) - load_barycentric_at_sample (ARB_gpu_shader5's interpolateAtSample()) - load_barycentric_at_offset (ARB_gpu_shader5's interpolateAtOffset()) Each of these take the interpolation mode (smooth or noperspective only) as a const_index, and produce a vec2. The last two also take a sample or offset source. We then introduce a new load_interpolated_input intrinsic, which is like a normal load_input intrinsic, but with an additional barycentric coordinate source. The intention is that flat inputs will still use regular load_input intrinsics. This makes them distinguishable from normal inputs that need fancy interpolation, while also providing all the necessary data. This nicely unifies regular inputs and interpolateAt functions. Qualifiers and variables become irrelevant; there are just load_barycentric intrinsics that determine the interpolation. v2: Document the interp_mode const_index value, define a new BARYCENTRIC() helper rather than using SYSTEM_VALUE() for some of them (requested by Jason Ekstrand). Signed-off-by: Kenneth Graunke <kenneth@whitecape.org> Reviewed-by: Chris Forbes <chrisforbes@google.com> Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
2016-07-12 09:46:43 +01:00
case nir_intrinsic_load_interpolated_input:
nir: Get rid of *_indirect variants of input/output load/store intrinsics There is some special-casing needed in a competent back-end. However, they can do their special-casing easily enough based on whether or not the offset is a constant. In the mean time, having the *_indirect variants adds special cases a number of places where they don't need to be and, in general, only complicates things. To complicate matters, NIR had no way to convdert an indirect load/store to a direct one in the case that the indirect was a constant so we would still not really get what the back-ends wanted. The best solution seems to be to get rid of the *_indirect variants entirely. This commit is a bunch of different changes squashed together: - nir: Get rid of *_indirect variants of input/output load/store intrinsics - nir/glsl: Stop handling UBO/SSBO load/stores differently depending on indirect - nir/lower_io: Get rid of load/store_foo_indirect - i965/fs: Get rid of load/store_foo_indirect - i965/vec4: Get rid of load/store_foo_indirect - tgsi_to_nir: Get rid of load/store_foo_indirect - ir3/nir: Use the new unified io intrinsics - vc4: Do all uniform loads with byte offsets - vc4/nir: Use the new unified io intrinsics - vc4: Fix load_user_clip_plane crash - vc4: add missing src for store outputs - vc4: Fix state uniforms - nir/lower_clip: Update to the new load/store intrinsics - nir/lower_two_sided_color: Update to the new load intrinsic NIR and i965 changes are Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> NIR indirect declarations and vc4 changes are Reviewed-by: Eric Anholt <eric@anholt.net> ir3 changes are Reviewed-by: Rob Clark <robdclark@gmail.com> NIR changes are Acked-by: Rob Clark <robdclark@gmail.com>
2015-11-25 22:14:05 +00:00
case nir_intrinsic_store_output:
return &instr->src[1];
case nir_intrinsic_store_ssbo:
nir: Get rid of *_indirect variants of input/output load/store intrinsics There is some special-casing needed in a competent back-end. However, they can do their special-casing easily enough based on whether or not the offset is a constant. In the mean time, having the *_indirect variants adds special cases a number of places where they don't need to be and, in general, only complicates things. To complicate matters, NIR had no way to convdert an indirect load/store to a direct one in the case that the indirect was a constant so we would still not really get what the back-ends wanted. The best solution seems to be to get rid of the *_indirect variants entirely. This commit is a bunch of different changes squashed together: - nir: Get rid of *_indirect variants of input/output load/store intrinsics - nir/glsl: Stop handling UBO/SSBO load/stores differently depending on indirect - nir/lower_io: Get rid of load/store_foo_indirect - i965/fs: Get rid of load/store_foo_indirect - i965/vec4: Get rid of load/store_foo_indirect - tgsi_to_nir: Get rid of load/store_foo_indirect - ir3/nir: Use the new unified io intrinsics - vc4: Do all uniform loads with byte offsets - vc4/nir: Use the new unified io intrinsics - vc4: Fix load_user_clip_plane crash - vc4: add missing src for store outputs - vc4: Fix state uniforms - nir/lower_clip: Update to the new load/store intrinsics - nir/lower_two_sided_color: Update to the new load intrinsic NIR and i965 changes are Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> NIR indirect declarations and vc4 changes are Reviewed-by: Eric Anholt <eric@anholt.net> ir3 changes are Reviewed-by: Rob Clark <robdclark@gmail.com> NIR changes are Acked-by: Rob Clark <robdclark@gmail.com>
2015-11-25 22:14:05 +00:00
case nir_intrinsic_store_per_vertex_output:
return &instr->src[2];
default:
return NULL;
}
}
/**
* Return the vertex index source for a load/store per_vertex intrinsic.
*/
nir_src *
nir_get_io_vertex_index_src(nir_intrinsic_instr *instr)
{
switch (instr->intrinsic) {
case nir_intrinsic_load_per_vertex_input:
case nir_intrinsic_load_per_vertex_output:
return &instr->src[0];
case nir_intrinsic_store_per_vertex_output:
return &instr->src[1];
default:
return NULL;
}
}