static bool
def_not_live_out(nir_ssa_def *def, void *state)
{
nir_block *after = state;
return !BITSET_TEST(after->live_in, def->live_index);
}
/**
* \note This routine refers to derived texture matrix values to
* compute the ENABLE_TEXMAT flags, but is only called on
* _NEW_TEXTURE_OBJECT/STATE. On changes to _NEW_TEXTURE_MATRIX,
* the ENABLE_TEXMAT flags are updated by _mesa_update_texture_matrices,
* above.
*
* \param ctx GL context.
*/
void
_mesa_update_texture_state(struct gl_context *ctx)
{
struct gl_program *prog[MESA_SHADER_STAGES];
int i;
int old_max_unit = ctx->Texture._MaxEnabledTexImageUnit;
BITSET_DECLARE(enabled_texture_units, MAX_COMBINED_TEXTURE_IMAGE_UNITS);
memcpy(prog, ctx->_Shader->CurrentProgram, sizeof(prog));
if (prog[MESA_SHADER_FRAGMENT] == NULL &&
_mesa_arb_fragment_program_enabled(ctx)) {
prog[MESA_SHADER_FRAGMENT] = ctx->FragmentProgram.Current;
}
/* TODO: only set this if there are actual changes */
ctx->NewState |= _NEW_TEXTURE_OBJECT | _NEW_TEXTURE_STATE;
ctx->Texture._GenFlags = 0x0;
ctx->Texture._TexMatEnabled = 0x0;
ctx->Texture._TexGenEnabled = 0x0;
ctx->Texture._MaxEnabledTexImageUnit = -1;
ctx->Texture._EnabledCoordUnits = 0x0;
memset(&enabled_texture_units, 0, sizeof(enabled_texture_units));
/* First, walk over our programs pulling in all the textures for them.
* Programs dictate specific texture targets to be enabled, and for a draw
* call to be valid they can't conflict about which texture targets are
* used.
*/
update_program_texture_state(ctx, prog, enabled_texture_units);
/* Also pull in any textures necessary for fixed function fragment shading.
*/
if (!prog[MESA_SHADER_FRAGMENT])
update_ff_texture_state(ctx, enabled_texture_units);
/* Now, clear out the _Current of any disabled texture units. */
for (i = 0; i <= ctx->Texture._MaxEnabledTexImageUnit; i++) {
if (!BITSET_TEST(enabled_texture_units, i))
_mesa_reference_texobj(&ctx->Texture.Unit[i]._Current, NULL);
}
for (i = ctx->Texture._MaxEnabledTexImageUnit + 1; i <= old_max_unit; i++) {
_mesa_reference_texobj(&ctx->Texture.Unit[i]._Current, NULL);
}
/* add fallback texture for SampleMapATI if there is nothing */
if (_mesa_ati_fragment_shader_enabled(ctx) &&
ctx->ATIFragmentShader.Current->Program)
fix_missing_textures_for_atifs(ctx,
ctx->ATIFragmentShader.Current->Program,
enabled_texture_units);
if (!prog[MESA_SHADER_FRAGMENT] || !prog[MESA_SHADER_VERTEX])
update_texgen(ctx);
}
/**
* \note This routine refers to derived texture matrix values to
* compute the ENABLE_TEXMAT flags, but is only called on
* _NEW_TEXTURE. On changes to _NEW_TEXTURE_MATRIX, the ENABLE_TEXMAT
* flags are updated by _mesa_update_texture_matrices, above.
*
* \param ctx GL context.
*/
static void
update_texture_state( struct gl_context *ctx )
{
struct gl_program *prog[MESA_SHADER_STAGES];
int i;
int old_max_unit = ctx->Texture._MaxEnabledTexImageUnit;
BITSET_DECLARE(enabled_texture_units, MAX_COMBINED_TEXTURE_IMAGE_UNITS);
for (i = 0; i < MESA_SHADER_STAGES; i++) {
if (ctx->_Shader->CurrentProgram[i] &&
ctx->_Shader->CurrentProgram[i]->LinkStatus) {
prog[i] = ctx->_Shader->CurrentProgram[i]->_LinkedShaders[i]->Program;
} else {
if (i == MESA_SHADER_FRAGMENT && ctx->FragmentProgram._Enabled)
prog[i] = &ctx->FragmentProgram.Current->Base;
else
prog[i] = NULL;
}
}
/* TODO: only set this if there are actual changes */
ctx->NewState |= _NEW_TEXTURE;
ctx->Texture._GenFlags = 0x0;
ctx->Texture._TexMatEnabled = 0x0;
ctx->Texture._TexGenEnabled = 0x0;
ctx->Texture._MaxEnabledTexImageUnit = -1;
ctx->Texture._EnabledCoordUnits = 0x0;
memset(&enabled_texture_units, 0, sizeof(enabled_texture_units));
/* First, walk over our programs pulling in all the textures for them.
* Programs dictate specific texture targets to be enabled, and for a draw
* call to be valid they can't conflict about which texture targets are
* used.
*/
update_program_texture_state(ctx, prog, enabled_texture_units);
/* Also pull in any textures necessary for fixed function fragment shading.
*/
if (!prog[MESA_SHADER_FRAGMENT])
update_ff_texture_state(ctx, enabled_texture_units);
/* Now, clear out the _Current of any disabled texture units. */
for (i = 0; i <= ctx->Texture._MaxEnabledTexImageUnit; i++) {
if (!BITSET_TEST(enabled_texture_units, i))
_mesa_reference_texobj(&ctx->Texture.Unit[i]._Current, NULL);
}
for (i = ctx->Texture._MaxEnabledTexImageUnit + 1; i <= old_max_unit; i++) {
_mesa_reference_texobj(&ctx->Texture.Unit[i]._Current, NULL);
}
if (!prog[MESA_SHADER_FRAGMENT] || !prog[MESA_SHADER_VERTEX])
update_texgen(ctx);
}
static void
qir_setup_use(struct vc4_compile *c, struct qblock *block, int ip,
struct qreg src)
{
int var = qir_reg_to_var(src);
if (var == -1)
return;
c->temp_start[var] = MIN2(c->temp_start[var], ip);
c->temp_end[var] = MAX2(c->temp_end[var], ip);
/* The use[] bitset marks when the block makes
* use of a variable without having completely
* defined that variable within the block.
*/
if (!BITSET_TEST(block->def, var))
BITSET_SET(block->use, var);
}
static void
validate_ssa_def(nir_ssa_def *def, validate_state *state)
{
assert(def->index < state->impl->ssa_alloc);
assert(!BITSET_TEST(state->ssa_defs_found, def->index));
BITSET_SET(state->ssa_defs_found, def->index);
assert(def->parent_instr == state->instr);
assert(def->num_components <= 4);
list_validate(&def->uses);
list_validate(&def->if_uses);
ssa_def_validate_state *def_state = ralloc(state->ssa_defs,
ssa_def_validate_state);
def_state->where_defined = state->impl;
def_state->uses = _mesa_set_create(def_state, _mesa_hash_pointer,
_mesa_key_pointer_equal);
def_state->if_uses = _mesa_set_create(def_state, _mesa_hash_pointer,
_mesa_key_pointer_equal);
_mesa_hash_table_insert(state->ssa_defs, def, def_state);
}
bool
fs_visitor::dead_code_eliminate()
{
bool progress = false;
calculate_live_intervals();
int num_vars = live_intervals->num_vars;
BITSET_WORD *live = ralloc_array(NULL, BITSET_WORD, BITSET_WORDS(num_vars));
foreach_block (block, cfg) {
memcpy(live, live_intervals->bd[block->num].liveout,
sizeof(BITSET_WORD) * BITSET_WORDS(num_vars));
foreach_inst_in_block_reverse(fs_inst, inst, block) {
if (inst->dst.file == GRF &&
!inst->has_side_effects() &&
!inst->writes_flag()) {
bool result_live = false;
if (inst->regs_written == 1) {
int var = live_intervals->var_from_reg(&inst->dst);
result_live = BITSET_TEST(live, var);
} else {
int var = live_intervals->var_from_vgrf[inst->dst.reg];
for (int i = 0; i < inst->regs_written; i++) {
result_live = result_live || BITSET_TEST(live, var + i);
}
}
if (!result_live) {
progress = true;
if (inst->writes_accumulator) {
inst->dst = fs_reg(retype(brw_null_reg(), inst->dst.type));
} else {
inst->opcode = BRW_OPCODE_NOP;
continue;
}
}
}
if (inst->dst.file == GRF) {
if (!inst->is_partial_write()) {
int var = live_intervals->var_from_vgrf[inst->dst.reg];
for (int i = 0; i < inst->regs_written; i++) {
BITSET_CLEAR(live, var + inst->dst.reg_offset + i);
}
}
}
for (int i = 0; i < inst->sources; i++) {
if (inst->src[i].file == GRF) {
int var = live_intervals->var_from_vgrf[inst->src[i].reg];
for (int j = 0; j < inst->regs_read(this, i); j++) {
BITSET_SET(live, var + inst->src[i].reg_offset + j);
}
}
}
}
}
static void
update_ff_texture_state(struct gl_context *ctx,
BITSET_WORD *enabled_texture_units)
{
int unit;
for (unit = 0; unit < ctx->Const.MaxTextureUnits; unit++) {
struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
GLbitfield mask;
bool complete;
if (texUnit->Enabled == 0x0)
continue;
/* If a shader already dictated what texture target was used for this
* unit, just go along with it.
*/
if (BITSET_TEST(enabled_texture_units, unit))
continue;
/* From the GL 4.4 compat specification, section 16.2 ("Texture Application"):
*
* "Texturing is enabled or disabled using the generic Enable and
* Disable commands, respectively, with the symbolic constants
* TEXTURE_1D, TEXTURE_2D, TEXTURE_RECTANGLE, TEXTURE_3D, or
* TEXTURE_CUBE_MAP to enable the one-, two-, rectangular,
* three-dimensional, or cube map texture, respectively. If more
* than one of these textures is enabled, the first one enabled
* from the following list is used:
*
* • cube map texture
* • three-dimensional texture
* • rectangular texture
* • two-dimensional texture
* • one-dimensional texture"
*
* Note that the TEXTURE_x_INDEX values are in high to low priority.
* Also:
*
* "If a texture unit is disabled or has an invalid or incomplete
* texture (as defined in section 8.17) bound to it, then blending
* is disabled for that texture unit. If the texture environment
* for a given enabled texture unit references a disabled texture
* unit, or an invalid or incomplete texture that is bound to
* another unit, then the results of texture blending are
* undefined."
*/
complete = false;
mask = texUnit->Enabled;
while (mask) {
const int texIndex = u_bit_scan(&mask);
struct gl_texture_object *texObj = texUnit->CurrentTex[texIndex];
struct gl_sampler_object *sampler = texUnit->Sampler ?
texUnit->Sampler : &texObj->Sampler;
if (!_mesa_is_texture_complete(texObj, sampler)) {
_mesa_test_texobj_completeness(ctx, texObj);
}
if (_mesa_is_texture_complete(texObj, sampler)) {
_mesa_reference_texobj(&texUnit->_Current, texObj);
complete = true;
break;
}
}
if (!complete)
continue;
/* if we get here, we know this texture unit is enabled */
BITSET_SET(enabled_texture_units, unit);
ctx->Texture._MaxEnabledTexImageUnit =
MAX2(ctx->Texture._MaxEnabledTexImageUnit, (int)unit);
ctx->Texture._EnabledCoordUnits |= 1 << unit;
update_tex_combine(ctx, texUnit);
}
}
static void
qir_setup_def(struct vc4_compile *c, struct qblock *block, int ip,
struct hash_table *partial_update_ht, struct qinst *inst)
{
/* The def[] bitset marks when an initialization in a
* block completely screens off previous updates of
* that variable.
*/
int var = qir_reg_to_var(inst->dst);
if (var == -1)
return;
c->temp_start[var] = MIN2(c->temp_start[var], ip);
c->temp_end[var] = MAX2(c->temp_end[var], ip);
/* If we've already tracked this as a def, or already used it within
* the block, there's nothing to do.
*/
if (BITSET_TEST(block->use, var) || BITSET_TEST(block->def, var))
return;
/* Easy, common case: unconditional full register update. */
if (inst->cond == QPU_COND_ALWAYS && !inst->dst.pack) {
BITSET_SET(block->def, var);
return;
}
/* Finally, look at the condition code and packing and mark it as a
* def. We need to make sure that we understand sequences
* instructions like:
*
* mov.zs t0, t1
* mov.zc t0, t2
*
* or:
*
* mmov t0.8a, t1
* mmov t0.8b, t2
* mmov t0.8c, t3
* mmov t0.8d, t4
*
* as defining the temp within the block, because otherwise dst's live
* range will get extended up the control flow to the top of the
* program.
*/
struct partial_update_state *state =
get_partial_update_state(partial_update_ht, inst);
uint8_t mask = qir_channels_written(inst);
if (inst->cond == QPU_COND_ALWAYS) {
state->channels |= mask;
} else {
for (int i = 0; i < 4; i++) {
if (!(mask & (1 << i)))
continue;
if (state->insts[i] &&
state->insts[i]->cond ==
qpu_cond_complement(inst->cond))
state->channels |= 1 << i;
else
state->insts[i] = inst;
}
}
if (state->channels == 0xf)
BITSET_SET(block->def, var);
}
