#include "brw_nir.h"
#include "brw_shader.h"
-#include "compiler/nir/glsl_to_nir.h"
+#include "common/gen_debug.h"
+#include "compiler/glsl_types.h"
#include "compiler/nir/nir_builder.h"
-#include "program/prog_to_nir.h"
static bool
is_input(nir_intrinsic_instr *intrin)
{
return intrin->intrinsic == nir_intrinsic_load_input ||
- intrin->intrinsic == nir_intrinsic_load_per_vertex_input;
+ intrin->intrinsic == nir_intrinsic_load_per_vertex_input ||
+ intrin->intrinsic == nir_intrinsic_load_interpolated_input;
}
static bool
* we don't know what part of a compound variable is accessed, we allocate
* storage for the entire thing.
*/
-struct add_const_offset_to_base_params {
- nir_builder b;
- nir_variable_mode mode;
-};
static bool
-add_const_offset_to_base_block(nir_block *block, void *closure)
+add_const_offset_to_base_block(nir_block *block, nir_builder *b,
+ nir_variable_mode mode)
{
- struct add_const_offset_to_base_params *params = closure;
- nir_builder *b = ¶ms->b;
-
- nir_foreach_instr_safe(block, instr) {
+ nir_foreach_instr_safe(instr, block) {
if (instr->type != nir_instr_type_intrinsic)
continue;
nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
- if ((params->mode == nir_var_shader_in && is_input(intrin)) ||
- (params->mode == nir_var_shader_out && is_output(intrin))) {
+ if ((mode == nir_var_shader_in && is_input(intrin)) ||
+ (mode == nir_var_shader_out && is_output(intrin))) {
nir_src *offset = nir_get_io_offset_src(intrin);
nir_const_value *const_offset = nir_src_as_const_value(*offset);
static void
add_const_offset_to_base(nir_shader *nir, nir_variable_mode mode)
{
- struct add_const_offset_to_base_params params = { .mode = mode };
-
- nir_foreach_function(nir, f) {
+ nir_foreach_function(f, nir) {
if (f->impl) {
- nir_builder_init(¶ms.b, f->impl);
- nir_foreach_block(f->impl, add_const_offset_to_base_block, ¶ms);
+ nir_builder b;
+ nir_builder_init(&b, f->impl);
+ nir_foreach_block(block, f->impl) {
+ add_const_offset_to_base_block(block, &b, mode);
+ }
}
}
}
static bool
-remap_vs_attrs(nir_block *block, void *closure)
+remap_vs_attrs(nir_block *block, shader_info *nir_info)
{
- GLbitfield64 inputs_read = *((GLbitfield64 *) closure);
-
- nir_foreach_instr(block, instr) {
+ nir_foreach_instr(instr, block) {
if (instr->type != nir_instr_type_intrinsic)
continue;
* before it and counting the bits.
*/
int attr = intrin->const_index[0];
- int slot = _mesa_bitcount_64(inputs_read & BITFIELD64_MASK(attr));
-
+ int slot = _mesa_bitcount_64(nir_info->inputs_read &
+ BITFIELD64_MASK(attr));
intrin->const_index[0] = 4 * slot;
}
}
}
static bool
-remap_inputs_with_vue_map(nir_block *block, void *closure)
+remap_inputs_with_vue_map(nir_block *block, const struct brw_vue_map *vue_map)
{
- const struct brw_vue_map *vue_map = closure;
-
- nir_foreach_instr(block, instr) {
+ nir_foreach_instr(instr, block) {
if (instr->type != nir_instr_type_intrinsic)
continue;
return true;
}
-struct remap_patch_urb_offsets_state {
- nir_builder b;
- const struct brw_vue_map *vue_map;
-};
+static bool
+remap_tess_levels(nir_builder *b, nir_intrinsic_instr *intr,
+ GLenum primitive_mode)
+{
+ const int location = nir_intrinsic_base(intr);
+ const unsigned component = nir_intrinsic_component(intr);
+ bool out_of_bounds;
+
+ if (location == VARYING_SLOT_TESS_LEVEL_INNER) {
+ switch (primitive_mode) {
+ case GL_QUADS:
+ /* gl_TessLevelInner[0..1] lives at DWords 3-2 (reversed). */
+ nir_intrinsic_set_base(intr, 0);
+ nir_intrinsic_set_component(intr, 3 - component);
+ out_of_bounds = false;
+ break;
+ case GL_TRIANGLES:
+ /* gl_TessLevelInner[0] lives at DWord 4. */
+ nir_intrinsic_set_base(intr, 1);
+ out_of_bounds = component > 0;
+ break;
+ case GL_ISOLINES:
+ out_of_bounds = true;
+ break;
+ default:
+ unreachable("Bogus tessellation domain");
+ }
+ } else if (location == VARYING_SLOT_TESS_LEVEL_OUTER) {
+ if (primitive_mode == GL_ISOLINES) {
+ /* gl_TessLevelOuter[0..1] lives at DWords 6-7 (in order). */
+ nir_intrinsic_set_base(intr, 1);
+ nir_intrinsic_set_component(intr, 2 + nir_intrinsic_component(intr));
+ out_of_bounds = component > 1;
+ } else {
+ /* Triangles use DWords 7-5 (reversed); Quads use 7-4 (reversed) */
+ nir_intrinsic_set_base(intr, 1);
+ nir_intrinsic_set_component(intr, 3 - nir_intrinsic_component(intr));
+ out_of_bounds = component == 3 && primitive_mode == GL_TRIANGLES;
+ }
+ } else {
+ return false;
+ }
+
+ if (out_of_bounds) {
+ if (nir_intrinsic_infos[intr->intrinsic].has_dest) {
+ b->cursor = nir_before_instr(&intr->instr);
+ nir_ssa_def *undef = nir_ssa_undef(b, 1, 32);
+ nir_ssa_def_rewrite_uses(&intr->dest.ssa, nir_src_for_ssa(undef));
+ }
+ nir_instr_remove(&intr->instr);
+ }
+
+ return true;
+}
static bool
-remap_patch_urb_offsets(nir_block *block, void *closure)
+remap_patch_urb_offsets(nir_block *block, nir_builder *b,
+ const struct brw_vue_map *vue_map,
+ GLenum tes_primitive_mode)
{
- struct remap_patch_urb_offsets_state *state = closure;
+ const bool is_passthrough_tcs = b->shader->info->name &&
+ strcmp(b->shader->info->name, "passthrough") == 0;
- nir_foreach_instr_safe(block, instr) {
+ nir_foreach_instr_safe(instr, block) {
if (instr->type != nir_instr_type_intrinsic)
continue;
nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
- gl_shader_stage stage = state->b.shader->stage;
+ gl_shader_stage stage = b->shader->stage;
if ((stage == MESA_SHADER_TESS_CTRL && is_output(intrin)) ||
(stage == MESA_SHADER_TESS_EVAL && is_input(intrin))) {
- int vue_slot = state->vue_map->varying_to_slot[intrin->const_index[0]];
+
+ if (!is_passthrough_tcs &&
+ remap_tess_levels(b, intrin, tes_primitive_mode))
+ continue;
+
+ int vue_slot = vue_map->varying_to_slot[intrin->const_index[0]];
assert(vue_slot != -1);
intrin->const_index[0] = vue_slot;
nir_const_value *const_vertex = nir_src_as_const_value(*vertex);
if (const_vertex) {
intrin->const_index[0] += const_vertex->u32[0] *
- state->vue_map->num_per_vertex_slots;
+ vue_map->num_per_vertex_slots;
} else {
- state->b.cursor = nir_before_instr(&intrin->instr);
+ b->cursor = nir_before_instr(&intrin->instr);
/* Multiply by the number of per-vertex slots. */
nir_ssa_def *vertex_offset =
- nir_imul(&state->b,
- nir_ssa_for_src(&state->b, *vertex, 1),
- nir_imm_int(&state->b,
- state->vue_map->num_per_vertex_slots));
+ nir_imul(b,
+ nir_ssa_for_src(b, *vertex, 1),
+ nir_imm_int(b,
+ vue_map->num_per_vertex_slots));
/* Add it to the existing offset */
nir_src *offset = nir_get_io_offset_src(intrin);
nir_ssa_def *total_offset =
- nir_iadd(&state->b, vertex_offset,
- nir_ssa_for_src(&state->b, *offset, 1));
+ nir_iadd(b, vertex_offset,
+ nir_ssa_for_src(b, *offset, 1));
nir_instr_rewrite_src(&intrin->instr, offset,
nir_src_for_ssa(total_offset));
void
brw_nir_lower_vs_inputs(nir_shader *nir,
- const struct brw_device_info *devinfo,
bool is_scalar,
bool use_legacy_snorm_formula,
const uint8_t *vs_attrib_wa_flags)
var->data.driver_location = var->data.location;
}
- /* Now use nir_lower_io to walk dereference chains. Attribute arrays
- * are loaded as one vec4 per element (or matrix column), so we use
- * type_size_vec4 here.
+ /* Now use nir_lower_io to walk dereference chains. Attribute arrays are
+ * loaded as one vec4 or dvec4 per element (or matrix column), depending on
+ * whether it is a double-precision type or not.
*/
- nir_lower_io(nir, nir_var_shader_in, type_size_vec4);
+ nir_lower_io(nir, nir_var_shader_in, type_size_vec4, 0);
/* This pass needs actual constants */
nir_opt_constant_folding(nir);
vs_attrib_wa_flags);
if (is_scalar) {
- /* Finally, translate VERT_ATTRIB_* values into the actual registers.
- *
- * Note that we can use nir->info.inputs_read instead of
- * key->inputs_read since the two are identical aside from Gen4-5
- * edge flag differences.
- */
- GLbitfield64 inputs_read = nir->info.inputs_read;
+ /* Finally, translate VERT_ATTRIB_* values into the actual registers. */
- nir_foreach_function(nir, function) {
+ nir_foreach_function(function, nir) {
if (function->impl) {
- nir_foreach_block(function->impl, remap_vs_attrs, &inputs_read);
+ nir_foreach_block(block, function->impl) {
+ remap_vs_attrs(block, nir->info);
+ }
}
}
}
}
/* Inputs are stored in vec4 slots, so use type_size_vec4(). */
- nir_lower_io(nir, nir_var_shader_in, type_size_vec4);
+ nir_lower_io(nir, nir_var_shader_in, type_size_vec4, 0);
if (is_scalar || nir->stage != MESA_SHADER_GEOMETRY) {
/* This pass needs actual constants */
add_const_offset_to_base(nir, nir_var_shader_in);
- nir_foreach_function(nir, function) {
+ nir_foreach_function(function, nir) {
if (function->impl) {
- nir_foreach_block(function->impl, remap_inputs_with_vue_map,
- (void *) vue_map);
+ nir_foreach_block(block, function->impl) {
+ remap_inputs_with_vue_map(block, vue_map);
+ }
}
}
}
void
brw_nir_lower_tes_inputs(nir_shader *nir, const struct brw_vue_map *vue_map)
{
- struct remap_patch_urb_offsets_state state;
- state.vue_map = vue_map;
-
foreach_list_typed(nir_variable, var, node, &nir->inputs) {
var->data.driver_location = var->data.location;
}
- nir_lower_io(nir, nir_var_shader_in, type_size_vec4);
+ nir_lower_io(nir, nir_var_shader_in, type_size_vec4, 0);
/* This pass needs actual constants */
nir_opt_constant_folding(nir);
add_const_offset_to_base(nir, nir_var_shader_in);
- nir_foreach_function(nir, function) {
+ nir_foreach_function(function, nir) {
if (function->impl) {
- nir_builder_init(&state.b, function->impl);
- nir_foreach_block(function->impl, remap_patch_urb_offsets, &state);
+ nir_builder b;
+ nir_builder_init(&b, function->impl);
+ nir_foreach_block(block, function->impl) {
+ remap_patch_urb_offsets(block, &b, vue_map,
+ nir->info->tess.primitive_mode);
+ }
}
}
}
void
-brw_nir_lower_fs_inputs(nir_shader *nir)
+brw_nir_lower_fs_inputs(nir_shader *nir,
+ const struct gen_device_info *devinfo,
+ const struct brw_wm_prog_key *key)
{
- nir_assign_var_locations(&nir->inputs, &nir->num_inputs, type_size_scalar);
- nir_lower_io(nir, nir_var_shader_in, type_size_scalar);
+ foreach_list_typed(nir_variable, var, node, &nir->inputs) {
+ var->data.driver_location = var->data.location;
+
+ /* Apply default interpolation mode.
+ *
+ * Everything defaults to smooth except for the legacy GL color
+ * built-in variables, which might be flat depending on API state.
+ */
+ if (var->data.interpolation == INTERP_MODE_NONE) {
+ const bool flat = key->flat_shade &&
+ (var->data.location == VARYING_SLOT_COL0 ||
+ var->data.location == VARYING_SLOT_COL1);
+
+ var->data.interpolation = flat ? INTERP_MODE_FLAT
+ : INTERP_MODE_SMOOTH;
+ }
+
+ /* On Ironlake and below, there is only one interpolation mode.
+ * Centroid interpolation doesn't mean anything on this hardware --
+ * there is no multisampling.
+ */
+ if (devinfo->gen < 6) {
+ var->data.centroid = false;
+ var->data.sample = false;
+ }
+ }
+
+ nir_lower_io_options lower_io_options = 0;
+ if (key->persample_interp)
+ lower_io_options |= nir_lower_io_force_sample_interpolation;
+
+ nir_lower_io(nir, nir_var_shader_in, type_size_vec4, lower_io_options);
+
+ /* This pass needs actual constants */
+ nir_opt_constant_folding(nir);
+
+ add_const_offset_to_base(nir, nir_var_shader_in);
}
void
brw_nir_lower_vue_outputs(nir_shader *nir,
bool is_scalar)
{
- if (is_scalar) {
- nir_assign_var_locations(&nir->outputs, &nir->num_outputs,
- type_size_vec4_times_4);
- nir_lower_io(nir, nir_var_shader_out, type_size_vec4_times_4);
- } else {
- nir_foreach_variable(var, &nir->outputs)
- var->data.driver_location = var->data.location;
- nir_lower_io(nir, nir_var_shader_out, type_size_vec4);
+ nir_foreach_variable(var, &nir->outputs) {
+ var->data.driver_location = var->data.location;
}
+
+ nir_lower_io(nir, nir_var_shader_out, type_size_vec4, 0);
}
void
-brw_nir_lower_tcs_outputs(nir_shader *nir, const struct brw_vue_map *vue_map)
+brw_nir_lower_tcs_outputs(nir_shader *nir, const struct brw_vue_map *vue_map,
+ GLenum tes_primitive_mode)
{
- struct remap_patch_urb_offsets_state state;
- state.vue_map = vue_map;
-
nir_foreach_variable(var, &nir->outputs) {
var->data.driver_location = var->data.location;
}
- nir_lower_io(nir, nir_var_shader_out, type_size_vec4);
+ nir_lower_io(nir, nir_var_shader_out, type_size_vec4, 0);
/* This pass needs actual constants */
nir_opt_constant_folding(nir);
add_const_offset_to_base(nir, nir_var_shader_out);
- nir_foreach_function(nir, function) {
+ nir_foreach_function(function, nir) {
if (function->impl) {
- nir_builder_init(&state.b, function->impl);
- nir_foreach_block(function->impl, remap_patch_urb_offsets, &state);
+ nir_builder b;
+ nir_builder_init(&b, function->impl);
+ nir_foreach_block(block, function->impl) {
+ remap_patch_urb_offsets(block, &b, vue_map, tes_primitive_mode);
+ }
}
}
}
void
brw_nir_lower_fs_outputs(nir_shader *nir)
{
- nir_assign_var_locations(&nir->outputs, &nir->num_outputs,
- type_size_scalar);
- nir_lower_io(nir, nir_var_shader_out, type_size_scalar);
-}
-
-static int
-type_size_scalar_bytes(const struct glsl_type *type)
-{
- return type_size_scalar(type) * 4;
-}
-
-static int
-type_size_vec4_bytes(const struct glsl_type *type)
-{
- return type_size_vec4(type) * 16;
-}
-
-static void
-brw_nir_lower_uniforms(nir_shader *nir, bool is_scalar)
-{
- if (is_scalar) {
- nir_assign_var_locations(&nir->uniforms, &nir->num_uniforms,
- type_size_scalar_bytes);
- nir_lower_io(nir, nir_var_uniform, type_size_scalar_bytes);
- } else {
- nir_assign_var_locations(&nir->uniforms, &nir->num_uniforms,
- type_size_vec4_bytes);
- nir_lower_io(nir, nir_var_uniform, type_size_vec4_bytes);
+ nir_foreach_variable(var, &nir->outputs) {
+ var->data.driver_location =
+ SET_FIELD(var->data.index, BRW_NIR_FRAG_OUTPUT_INDEX) |
+ SET_FIELD(var->data.location, BRW_NIR_FRAG_OUTPUT_LOCATION);
}
+
+ nir_lower_io(nir, nir_var_shader_out, type_size_dvec4, 0);
}
void
{
nir_assign_var_locations(&nir->shared, &nir->num_shared,
type_size_scalar_bytes);
- nir_lower_io(nir, nir_var_shared, type_size_scalar_bytes);
+ nir_lower_io(nir, nir_var_shared, type_size_scalar_bytes, 0);
}
#define OPT(pass, ...) ({ \
#define OPT_V(pass, ...) NIR_PASS_V(nir, pass, ##__VA_ARGS__)
static nir_shader *
-nir_optimize(nir_shader *nir, bool is_scalar)
+nir_optimize(nir_shader *nir, const struct brw_compiler *compiler,
+ bool is_scalar)
{
+ nir_variable_mode indirect_mask = 0;
+ if (compiler->glsl_compiler_options[nir->stage].EmitNoIndirectInput)
+ indirect_mask |= nir_var_shader_in;
+ if (compiler->glsl_compiler_options[nir->stage].EmitNoIndirectOutput)
+ indirect_mask |= nir_var_shader_out;
+ if (compiler->glsl_compiler_options[nir->stage].EmitNoIndirectTemp)
+ indirect_mask |= nir_var_local;
+
bool progress;
do {
progress = false;
OPT_V(nir_lower_vars_to_ssa);
+ OPT(nir_opt_copy_prop_vars);
if (is_scalar) {
- OPT_V(nir_lower_alu_to_scalar);
+ OPT(nir_lower_alu_to_scalar);
}
OPT(nir_copy_prop);
if (is_scalar) {
- OPT_V(nir_lower_phis_to_scalar);
+ OPT(nir_lower_phis_to_scalar);
}
OPT(nir_copy_prop);
OPT(nir_opt_dce);
OPT(nir_opt_cse);
- OPT(nir_opt_peephole_select);
+ OPT(nir_opt_peephole_select, 0);
OPT(nir_opt_algebraic);
OPT(nir_opt_constant_folding);
OPT(nir_opt_dead_cf);
+ if (OPT(nir_opt_trivial_continues)) {
+ /* If nir_opt_trivial_continues makes progress, then we need to clean
+ * things up if we want any hope of nir_opt_if or nir_opt_loop_unroll
+ * to make progress.
+ */
+ OPT(nir_copy_prop);
+ OPT(nir_opt_dce);
+ }
+ OPT(nir_opt_if);
+ if (nir->options->max_unroll_iterations != 0) {
+ OPT(nir_opt_loop_unroll, indirect_mask);
+ }
OPT(nir_opt_remove_phis);
OPT(nir_opt_undef);
+ OPT_V(nir_lower_doubles, nir_lower_drcp |
+ nir_lower_dsqrt |
+ nir_lower_drsq |
+ nir_lower_dtrunc |
+ nir_lower_dfloor |
+ nir_lower_dceil |
+ nir_lower_dfract |
+ nir_lower_dround_even |
+ nir_lower_dmod);
+ OPT_V(nir_lower_64bit_pack);
} while (progress);
return nir;
* is_scalar = true to scalarize everything prior to code gen.
*/
nir_shader *
-brw_preprocess_nir(nir_shader *nir, bool is_scalar)
+brw_preprocess_nir(const struct brw_compiler *compiler, nir_shader *nir)
{
+ const struct gen_device_info *devinfo = compiler->devinfo;
bool progress; /* Written by OPT and OPT_V */
(void)progress;
+ const bool is_scalar = compiler->scalar_stage[nir->stage];
+
if (nir->stage == MESA_SHADER_GEOMETRY)
OPT(nir_lower_gs_intrinsics);
+ /* See also brw_nir_trig_workarounds.py */
+ if (compiler->precise_trig &&
+ !(devinfo->gen >= 10 || devinfo->is_kabylake))
+ OPT(brw_nir_apply_trig_workarounds);
+
static const nir_lower_tex_options tex_options = {
.lower_txp = ~0,
+ .lower_txf_offset = true,
+ .lower_rect_offset = true,
+ .lower_txd_cube_map = true,
};
OPT(nir_lower_tex, &tex_options);
OPT(nir_split_var_copies);
- nir = nir_optimize(nir, is_scalar);
+ nir = nir_optimize(nir, compiler, is_scalar);
if (is_scalar) {
OPT_V(nir_lower_load_const_to_scalar);
/* Lower a bunch of stuff */
OPT_V(nir_lower_var_copies);
+ OPT_V(nir_lower_clip_cull_distance_arrays);
+
+ nir_variable_mode indirect_mask = 0;
+ if (compiler->glsl_compiler_options[nir->stage].EmitNoIndirectInput)
+ indirect_mask |= nir_var_shader_in;
+ if (compiler->glsl_compiler_options[nir->stage].EmitNoIndirectOutput)
+ indirect_mask |= nir_var_shader_out;
+ if (compiler->glsl_compiler_options[nir->stage].EmitNoIndirectTemp)
+ indirect_mask |= nir_var_local;
+
+ nir_lower_indirect_derefs(nir, indirect_mask);
+
+ nir_lower_int64(nir, nir_lower_imul64 |
+ nir_lower_isign64 |
+ nir_lower_divmod64);
+
/* Get rid of split copies */
- nir = nir_optimize(nir, is_scalar);
+ nir = nir_optimize(nir, compiler, is_scalar);
- OPT(nir_remove_dead_variables);
+ OPT(nir_remove_dead_variables, nir_var_local);
return nir;
}
* will not work.
*/
nir_shader *
-brw_postprocess_nir(nir_shader *nir,
- const struct brw_device_info *devinfo,
+brw_postprocess_nir(nir_shader *nir, const struct brw_compiler *compiler,
bool is_scalar)
{
+ const struct gen_device_info *devinfo = compiler->devinfo;
bool debug_enabled =
(INTEL_DEBUG & intel_debug_flag_for_shader_stage(nir->stage));
bool progress; /* Written by OPT and OPT_V */
(void)progress;
- nir = nir_optimize(nir, is_scalar);
+ nir = nir_optimize(nir, compiler, is_scalar);
if (devinfo->gen >= 6) {
/* Try and fuse multiply-adds */
OPT(nir_opt_algebraic_late);
- OPT(nir_lower_locals_to_regs);
-
OPT_V(nir_lower_to_source_mods);
OPT(nir_copy_prop);
OPT(nir_opt_dce);
+ OPT(nir_opt_move_comparisons);
+
+ OPT(nir_lower_locals_to_regs);
if (unlikely(debug_enabled)) {
/* Re-index SSA defs so we print more sensible numbers. */
- nir_foreach_function(nir, function) {
+ nir_foreach_function(function, nir) {
if (function->impl)
nir_index_ssa_defs(function->impl);
}
return nir;
}
-nir_shader *
-brw_create_nir(struct brw_context *brw,
- const struct gl_shader_program *shader_prog,
- const struct gl_program *prog,
- gl_shader_stage stage,
- bool is_scalar)
-{
- struct gl_context *ctx = &brw->ctx;
- const nir_shader_compiler_options *options =
- ctx->Const.ShaderCompilerOptions[stage].NirOptions;
- bool progress;
- nir_shader *nir;
-
- /* First, lower the GLSL IR or Mesa IR to NIR */
- if (shader_prog) {
- nir = glsl_to_nir(shader_prog, stage, options);
- } else {
- nir = prog_to_nir(prog, options);
- OPT_V(nir_convert_to_ssa); /* turn registers into SSA */
- }
- nir_validate_shader(nir);
-
- (void)progress;
-
- nir = brw_preprocess_nir(nir, is_scalar);
-
- OPT(nir_lower_system_values);
- OPT_V(brw_nir_lower_uniforms, is_scalar);
-
- if (shader_prog) {
- OPT_V(nir_lower_samplers, shader_prog);
- OPT_V(nir_lower_atomics, shader_prog);
- }
-
- return nir;
-}
-
nir_shader *
brw_nir_apply_sampler_key(nir_shader *nir,
- const struct brw_device_info *devinfo,
+ const struct brw_compiler *compiler,
const struct brw_sampler_prog_key_data *key_tex,
bool is_scalar)
{
+ const struct gen_device_info *devinfo = compiler->devinfo;
nir_lower_tex_options tex_options = { 0 };
/* Iron Lake and prior require lowering of all rectangle textures */
tex_options.swizzles[s][c] = GET_SWZ(key_tex->swizzles[s], c);
}
+ /* Prior to Haswell, we have to lower gradients on shadow samplers */
+ tex_options.lower_txd_shadow = devinfo->gen < 8 && !devinfo->is_haswell;
+
+ tex_options.lower_y_uv_external = key_tex->y_uv_image_mask;
+ tex_options.lower_y_u_v_external = key_tex->y_u_v_image_mask;
+ tex_options.lower_yx_xuxv_external = key_tex->yx_xuxv_image_mask;
+
if (nir_lower_tex(nir, &tex_options)) {
nir_validate_shader(nir);
- nir = nir_optimize(nir, is_scalar);
+ nir = nir_optimize(nir, compiler, is_scalar);
}
return nir;
}
enum brw_reg_type
-brw_type_for_nir_type(nir_alu_type type)
+brw_type_for_nir_type(const struct gen_device_info *devinfo, nir_alu_type type)
{
switch (type) {
case nir_type_uint:
case nir_type_float64:
return BRW_REGISTER_TYPE_DF;
case nir_type_int64:
+ return devinfo->gen < 8 ? BRW_REGISTER_TYPE_DF : BRW_REGISTER_TYPE_Q;
case nir_type_uint64:
- /* TODO we should only see these in moves, so for now it's ok, but when
- * we add actual 64-bit integer support we should fix this.
- */
- return BRW_REGISTER_TYPE_DF;
+ return devinfo->gen < 8 ? BRW_REGISTER_TYPE_DF : BRW_REGISTER_TYPE_UQ;
default:
unreachable("unknown type");
}
projects / mesa.git / blobdiff