#include "vc4_context.h"
#include "vc4_qpu.h"
#include "vc4_qir.h"
-#include "mesa/state_tracker/st_glsl_types.h"
static struct qreg
ntq_get_src(struct vc4_compile *c, nir_src src, int i);
ntq_emit_cf_list(struct vc4_compile *c, struct exec_list *list);
static int
-type_size(const struct glsl_type *type)
+type_size(const struct glsl_type *type, bool bindless)
{
return glsl_count_attribute_slots(type, false);
}
-static int
-uniforms_type_size(const struct glsl_type *type)
-{
- return st_glsl_storage_type_size(type, false);
-}
-
static void
resize_qreg_array(struct vc4_compile *c,
struct qreg **regs,
indirect_uniform_load(struct vc4_compile *c, nir_intrinsic_instr *intr)
{
struct qreg indirect_offset = ntq_get_src(c, intr->src[0], 0);
- uint32_t offset = nir_intrinsic_base(intr);
- struct vc4_compiler_ubo_range *range = NULL;
- unsigned i;
- for (i = 0; i < c->num_uniform_ranges; i++) {
- range = &c->ubo_ranges[i];
- if (offset >= range->src_offset &&
- offset < range->src_offset + range->size) {
- break;
- }
- }
- /* The driver-location-based offset always has to be within a declared
- * uniform range.
- */
- assert(range);
- if (!range->used) {
- range->used = true;
- range->dst_offset = c->next_ubo_dst_offset;
- c->next_ubo_dst_offset += range->size;
- c->num_ubo_ranges++;
- }
-
- offset -= range->src_offset;
-
- /* Adjust for where we stored the TGSI register base. */
- indirect_offset = qir_ADD(c, indirect_offset,
- qir_uniform_ui(c, (range->dst_offset +
- offset)));
/* Clamp to [0, array size). Note that MIN/MAX are signed. */
+ uint32_t range = nir_intrinsic_range(intr);
indirect_offset = qir_MAX(c, indirect_offset, qir_uniform_ui(c, 0));
indirect_offset = qir_MIN_NOIMM(c, indirect_offset,
- qir_uniform_ui(c, (range->dst_offset +
- range->size - 4)));
+ qir_uniform_ui(c, range - 4));
qir_ADD_dest(c, qir_reg(QFILE_TEX_S_DIRECT, 0),
indirect_offset,
- qir_uniform(c, QUNIFORM_UBO_ADDR, 0));
+ qir_uniform(c, QUNIFORM_UBO0_ADDR,
+ nir_intrinsic_base(intr)));
c->num_texture_samples++;
static struct qreg
vc4_ubo_load(struct vc4_compile *c, nir_intrinsic_instr *intr)
{
- unsigned buffer_index = nir_src_as_uint(intr->src[0]);
+ int buffer_index = nir_src_as_uint(intr->src[0]);
assert(buffer_index == 1);
assert(c->stage == QSTAGE_FRAG);
qir_ADD_dest(c, qir_reg(QFILE_TEX_S_DIRECT, 0),
offset,
- qir_uniform(c, QUNIFORM_UBO_ADDR, buffer_index));
+ qir_uniform(c, QUNIFORM_UBO1_ADDR, 0));
c->num_texture_samples++;
c->output_slots[decl_offset].swizzle = swizzle;
}
-static void
-declare_uniform_range(struct vc4_compile *c, uint32_t start, uint32_t size)
-{
- unsigned array_id = c->num_uniform_ranges++;
- if (array_id >= c->ubo_ranges_array_size) {
- c->ubo_ranges_array_size = MAX2(c->ubo_ranges_array_size * 2,
- array_id + 1);
- c->ubo_ranges = reralloc(c, c->ubo_ranges,
- struct vc4_compiler_ubo_range,
- c->ubo_ranges_array_size);
- }
-
- c->ubo_ranges[array_id].dst_offset = 0;
- c->ubo_ranges[array_id].src_offset = start;
- c->ubo_ranges[array_id].size = size;
- c->ubo_ranges[array_id].used = false;
-}
-
static bool
ntq_src_is_only_ssa_def_user(nir_src *src)
{
struct qreg result;
switch (instr->op) {
- case nir_op_fmov:
- case nir_op_imov:
+ case nir_op_mov:
result = qir_MOV(c, src[0]);
break;
case nir_op_fmul:
else
point_size = qir_uniform_f(c, 1.0);
- /* Workaround: HW-2726 PTB does not handle zero-size points (BCM2835,
- * BCM21553).
- */
- point_size = qir_FMAX(c, point_size, qir_uniform_f(c, .125));
-
qir_VPM_WRITE(c, point_size);
}
vc4_optimize_nir(struct nir_shader *s)
{
bool progress;
+ unsigned lower_flrp =
+ (s->options->lower_flrp16 ? 16 : 0) |
+ (s->options->lower_flrp32 ? 32 : 0) |
+ (s->options->lower_flrp64 ? 64 : 0);
do {
progress = false;
NIR_PASS_V(s, nir_lower_vars_to_ssa);
- NIR_PASS(progress, s, nir_lower_alu_to_scalar);
+ NIR_PASS(progress, s, nir_lower_alu_to_scalar, NULL, NULL);
NIR_PASS(progress, s, nir_lower_phis_to_scalar);
NIR_PASS(progress, s, nir_copy_prop);
NIR_PASS(progress, s, nir_opt_remove_phis);
NIR_PASS(progress, s, nir_opt_peephole_select, 8, true, true);
NIR_PASS(progress, s, nir_opt_algebraic);
NIR_PASS(progress, s, nir_opt_constant_folding);
+ if (lower_flrp != 0) {
+ bool lower_flrp_progress = false;
+
+ NIR_PASS(lower_flrp_progress, s, nir_lower_flrp,
+ lower_flrp,
+ false /* always_precise */,
+ s->options->lower_ffma);
+ if (lower_flrp_progress) {
+ NIR_PASS(progress, s, nir_opt_constant_folding);
+ progress = true;
+ }
+
+ /* Nothing should rematerialize any flrps, so we only
+ * need to do this lowering once.
+ */
+ lower_flrp = 0;
+ }
+
NIR_PASS(progress, s, nir_opt_undef);
NIR_PASS(progress, s, nir_opt_loop_unroll,
nir_var_shader_in |
}
}
-static void
-ntq_setup_uniforms(struct vc4_compile *c)
-{
- nir_foreach_variable(var, &c->s->uniforms) {
- uint32_t vec4_count = uniforms_type_size(var->type);
- unsigned vec4_size = 4 * sizeof(float);
-
- declare_uniform_range(c, var->data.driver_location * vec4_size,
- vec4_count * vec4_size);
-
- }
-}
-
/**
* Sets up the mapping from nir_register to struct qreg *.
*
{
struct qreg *qregs = ntq_init_ssa_def(c, &instr->def);
for (int i = 0; i < instr->def.num_components; i++)
- qregs[i] = qir_uniform_ui(c, instr->value.u32[i]);
+ qregs[i] = qir_uniform_ui(c, instr->value[i].u32);
_mesa_hash_table_insert(c->def_ht, &instr->def, qregs);
}
0));
break;
- case nir_intrinsic_load_alpha_ref_float:
- ntq_store_dest(c, &instr->dest, 0,
- qir_uniform(c, QUNIFORM_ALPHA_REF, 0));
- break;
-
case nir_intrinsic_load_sample_mask_in:
ntq_store_dest(c, &instr->dest, 0,
qir_uniform(c, QUNIFORM_SAMPLE_MASK, 0));
ntq_setup_inputs(c);
ntq_setup_outputs(c);
- ntq_setup_uniforms(c);
/* Find the main function and emit the body. */
nir_foreach_function(function, c->s) {
.lower_fdiv = true,
.lower_ffma = true,
.lower_flrp32 = true,
+ .lower_fmod = true,
.lower_fpow = true,
.lower_fsat = true,
.lower_fsqrt = true,
.lower_ldexp = true,
.lower_negate = true,
- .native_integers = true,
+ .lower_rotate = true,
+ .lower_to_scalar = true,
.max_unroll_iterations = 32,
};
c->s = nir_shader_clone(c, key->shader_state->base.ir.nir);
- if (stage == QSTAGE_FRAG) {
- if (c->fs_key->alpha_test_func != COMPARE_FUNC_ALWAYS) {
- NIR_PASS_V(c->s, nir_lower_alpha_test,
- c->fs_key->alpha_test_func,
- c->fs_key->sample_alpha_to_one &&
- c->fs_key->msaa);
- }
+ if (stage == QSTAGE_FRAG)
NIR_PASS_V(c->s, vc4_nir_lower_blend, c);
- }
struct nir_lower_tex_options tex_options = {
/* We would need to implement txs, but we don't want the
if (c->key->ucp_enables) {
if (stage == QSTAGE_FRAG) {
- NIR_PASS_V(c->s, nir_lower_clip_fs, c->key->ucp_enables);
+ NIR_PASS_V(c->s, nir_lower_clip_fs,
+ c->key->ucp_enables, false);
} else {
NIR_PASS_V(c->s, nir_lower_clip_vs,
- c->key->ucp_enables, false);
+ c->key->ucp_enables, false, false, NULL);
NIR_PASS_V(c->s, nir_lower_io_to_scalar,
nir_var_shader_out);
}
vc4_optimize_nir(c->s);
+ /* Do late algebraic optimization to turn add(a, neg(b)) back into
+ * subs, then the mandatory cleanup after algebraic. Note that it may
+ * produce fnegs, and if so then we need to keep running to squash
+ * fneg(fneg(a)).
+ */
+ bool more_late_algebraic = true;
+ while (more_late_algebraic) {
+ more_late_algebraic = false;
+ NIR_PASS(more_late_algebraic, c->s, nir_opt_algebraic_late);
+ NIR_PASS_V(c->s, nir_opt_constant_folding);
+ NIR_PASS_V(c->s, nir_copy_prop);
+ NIR_PASS_V(c->s, nir_opt_dce);
+ NIR_PASS_V(c->s, nir_opt_cse);
+ }
+
NIR_PASS_V(c->s, nir_lower_bool_to_int32);
NIR_PASS_V(c->s, nir_convert_from_ssa, true);
s = tgsi_to_nir(cso->tokens, pctx->screen);
}
+ if (s->info.stage == MESA_SHADER_VERTEX)
+ NIR_PASS_V(s, nir_lower_point_size, 1.0f, 0.0f);
+
NIR_PASS_V(s, nir_lower_io, nir_var_all, type_size,
(nir_lower_io_options)0);
shader->fs_threaded = c->fs_threaded;
- /* Copy the compiler UBO range state to the compiled shader, dropping
- * out arrays that were never referenced by an indirect load.
- *
- * (Note that QIR dead code elimination of an array access still
- * leaves that array alive, though)
- */
- if (c->num_ubo_ranges) {
- shader->num_ubo_ranges = c->num_ubo_ranges;
- shader->ubo_ranges = ralloc_array(shader, struct vc4_ubo_range,
- c->num_ubo_ranges);
- uint32_t j = 0;
- for (int i = 0; i < c->num_uniform_ranges; i++) {
- struct vc4_compiler_ubo_range *range =
- &c->ubo_ranges[i];
- if (!range->used)
- continue;
-
- shader->ubo_ranges[j].dst_offset = range->dst_offset;
- shader->ubo_ranges[j].src_offset = range->src_offset;
- shader->ubo_ranges[j].size = range->size;
- shader->ubo_size += c->ubo_ranges[i].size;
- j++;
- }
- }
- if (shader->ubo_size) {
- if (vc4_debug & VC4_DEBUG_SHADERDB) {
- fprintf(stderr, "SHADER-DB: %s prog %d/%d: %d UBO uniforms\n",
- qir_get_stage_name(c->stage),
- c->program_id, c->variant_id,
- shader->ubo_size / 4);
- }
- }
-
if ((vc4_debug & VC4_DEBUG_SHADERDB) && stage == QSTAGE_FRAG) {
fprintf(stderr, "SHADER-DB: %s prog %d/%d: %d FS threads\n",
qir_get_stage_name(c->stage),