p0_unpacked.bias_supplied = true;
break;
case nir_tex_src_lod:
- /* XXX: Needs base level addition */
coords[next_coord++] =
- ntq_get_src(c, instr->src[i].src, 0);
+ vir_FADD(c,
+ ntq_get_src(c, instr->src[i].src, 0),
+ vir_uniform(c, QUNIFORM_TEXTURE_FIRST_LEVEL,
+ unit));
if (instr->op != nir_texop_txf &&
instr->op != nir_texop_tg4) {
STATIC_ASSERT(PIPE_SWIZZLE_X == 0);
chan = return_values[i / 2];
- enum v3d_qpu_input_unpack unpack;
- if (i & 1)
- unpack = V3D_QPU_UNPACK_H;
- else
- unpack = V3D_QPU_UNPACK_L;
+ if (nir_alu_type_get_base_type(instr->dest_type) ==
+ nir_type_float) {
+ enum v3d_qpu_input_unpack unpack;
+ if (i & 1)
+ unpack = V3D_QPU_UNPACK_H;
+ else
+ unpack = V3D_QPU_UNPACK_L;
- chan = vir_FMOV(c, chan);
- vir_set_unpack(c->defs[chan.index], 0, unpack);
+ chan = vir_FMOV(c, chan);
+ vir_set_unpack(c->defs[chan.index], 0, unpack);
+ } else {
+ /* If we're unpacking the low field, shift it
+ * up to the top first.
+ */
+ if ((i & 1) == 0) {
+ chan = vir_SHL(c, chan,
+ vir_uniform_ui(c, 16));
+ }
+
+ /* Do proper sign extension to a 32-bit int. */
+ if (nir_alu_type_get_base_type(instr->dest_type) ==
+ nir_type_int) {
+ chan = vir_ASR(c, chan,
+ vir_uniform_ui(c, 16));
+ } else {
+ chan = vir_SHR(c, chan,
+ vir_uniform_ui(c, 16));
+ }
+ }
} else {
chan = vir_MOV(c, return_values[i]);
}
emit_fragment_input(struct v3d_compile *c, int attr, nir_variable *var)
{
for (int i = 0; i < glsl_get_vector_elements(var->type); i++) {
- c->inputs[attr * 4 + i] =
- emit_fragment_varying(c, var, i);
+ int chan = var->data.location_frac + i;
+ c->inputs[attr * 4 + chan] =
+ emit_fragment_varying(c, var, chan);
}
}
case nir_op_imov:
result = vir_MOV(c, src[0]);
break;
+
+ case nir_op_fneg:
+ result = vir_XOR(c, src[0], vir_uniform_ui(c, 1 << 31));
+ break;
+ case nir_op_ineg:
+ result = vir_NEG(c, src[0]);
+ break;
+
case nir_op_fmul:
result = vir_FMUL(c, src[0], src[1]);
break;
static void
emit_frag_end(struct v3d_compile *c)
{
- uint32_t discard_cond = V3D_QPU_COND_NONE;
- if (c->s->info.fs.uses_discard) {
- vir_PF(c, vir_MOV(c, c->discard), V3D_QPU_PF_PUSHZ);
- discard_cond = V3D_QPU_COND_IFA;
- }
-
/* XXX
if (c->output_sample_mask_index != -1) {
vir_MS_MASK(c, c->outputs[c->output_sample_mask_index]);
}
*/
+ bool has_any_tlb_color_write = false;
+ for (int rt = 0; rt < c->fs_key->nr_cbufs; rt++) {
+ if (c->output_color_var[rt])
+ has_any_tlb_color_write = true;
+ }
+
if (c->output_position_index != -1) {
struct qinst *inst = vir_MOV_dest(c,
vir_reg(QFILE_TLBU, 0),
c->outputs[c->output_position_index]);
- vir_set_cond(inst, discard_cond);
inst->src[vir_get_implicit_uniform_src(inst)] =
vir_uniform_ui(c,
TLB_TYPE_DEPTH |
TLB_DEPTH_TYPE_PER_PIXEL |
0xffffff00);
- } else if (c->s->info.fs.uses_discard) {
+ } else if (c->s->info.fs.uses_discard || !has_any_tlb_color_write) {
+ /* Emit passthrough Z if it needed to be delayed until shader
+ * end due to potential discards.
+ *
+ * Since (single-threaded) fragment shaders always need a TLB
+ * write, emit passthrouh Z if we didn't have any color
+ * buffers and flag us as potentially discarding, so that we
+ * can use Z as the TLB write.
+ */
+ c->s->info.fs.uses_discard = true;
+
struct qinst *inst = vir_MOV_dest(c,
vir_reg(QFILE_TLBU, 0),
vir_reg(QFILE_NULL, 0));
- vir_set_cond(inst, discard_cond);
inst->src[vir_get_implicit_uniform_src(inst)] =
vir_uniform_ui(c,
* uniform setup
*/
- if (c->output_color_var) {
- nir_variable *var = c->output_color_var;
+ for (int rt = 0; rt < c->fs_key->nr_cbufs; rt++) {
+ if (!c->output_color_var[rt])
+ continue;
+
+ nir_variable *var = c->output_color_var[rt];
struct qreg *color = &c->outputs[var->data.driver_location * 4];
int num_components = glsl_get_vector_elements(var->type);
- uint32_t conf = ~0;
+ uint32_t conf = 0xffffff00;
struct qinst *inst;
+ conf |= TLB_SAMPLE_MODE_PER_PIXEL;
+ conf |= (7 - rt) << TLB_RENDER_TARGET_SHIFT;
+
assert(num_components != 0);
switch (glsl_get_base_type(var->type)) {
case GLSL_TYPE_UINT:
case GLSL_TYPE_INT:
- conf = (TLB_TYPE_I32_COLOR |
- TLB_SAMPLE_MODE_PER_PIXEL |
- ((7 - 0) << TLB_RENDER_TARGET_SHIFT) |
- ((num_components - 1) <<
- TLB_VEC_SIZE_MINUS_1_SHIFT) |
- 0xffffff00);
-
+ conf |= TLB_TYPE_I32_COLOR;
+ conf |= ((num_components - 1) <<
+ TLB_VEC_SIZE_MINUS_1_SHIFT);
inst = vir_MOV_dest(c, vir_reg(QFILE_TLBU, 0), color[0]);
- vir_set_cond(inst, discard_cond);
inst->src[vir_get_implicit_uniform_src(inst)] =
vir_uniform_ui(c, conf);
for (int i = 1; i < num_components; i++) {
inst = vir_MOV_dest(c, vir_reg(QFILE_TLB, 0),
color[i]);
- vir_set_cond(inst, discard_cond);
}
break;
struct qreg b = color[2];
struct qreg a = color[3];
- if (c->fs_key->swap_color_rb) {
+ if (c->fs_key->f32_color_rb) {
+ conf |= TLB_TYPE_F32_COLOR;
+ conf |= ((num_components - 1) <<
+ TLB_VEC_SIZE_MINUS_1_SHIFT);
+ } else {
+ conf |= TLB_TYPE_F16_COLOR;
+ conf |= TLB_F16_SWAP_HI_LO;
+ if (num_components >= 3)
+ conf |= TLB_VEC_SIZE_4_F16;
+ else
+ conf |= TLB_VEC_SIZE_2_F16;
+ }
+
+ if (c->fs_key->swap_color_rb & (1 << rt)) {
r = color[2];
b = color[0];
}
- inst = vir_VFPACK_dest(c, vir_reg(QFILE_TLB, 0), r, g);
- vir_set_cond(inst, discard_cond);
- inst = vir_VFPACK_dest(c, vir_reg(QFILE_TLB, 0), b, a);
- vir_set_cond(inst, discard_cond);
+ if (c->fs_key->f32_color_rb & (1 << rt)) {
+ inst = vir_MOV_dest(c, vir_reg(QFILE_TLBU, 0), color[0]);
+ inst->src[vir_get_implicit_uniform_src(inst)] =
+ vir_uniform_ui(c, conf);
+
+ for (int i = 1; i < num_components; i++) {
+ inst = vir_MOV_dest(c, vir_reg(QFILE_TLB, 0),
+ color[i]);
+ }
+ } else {
+ inst = vir_VFPACK_dest(c, vir_reg(QFILE_TLB, 0), r, g);
+ if (conf != ~0) {
+ inst->dst.file = QFILE_TLBU;
+ inst->src[vir_get_implicit_uniform_src(inst)] =
+ vir_uniform_ui(c, conf);
+ }
+
+ inst = vir_VFPACK_dest(c, vir_reg(QFILE_TLB, 0), b, a);
+ }
break;
}
}
qsort(&vars, num_entries, sizeof(*vars), driver_location_compare);
uint32_t vpm_components_queued = 0;
- if (c->s->stage == MESA_SHADER_VERTEX) {
+ if (c->s->info.stage == MESA_SHADER_VERTEX) {
bool uses_iid = c->s->info.system_values_read &
(1ull << SYSTEM_VALUE_INSTANCE_ID);
bool uses_vid = c->s->info.system_values_read &
resize_qreg_array(c, &c->inputs, &c->inputs_array_size,
(loc + 1) * 4);
- if (c->s->stage == MESA_SHADER_FRAGMENT) {
+ if (c->s->info.stage == MESA_SHADER_FRAGMENT) {
if (var->data.location == VARYING_SLOT_POS) {
emit_fragcoord_input(c, loc);
} else if (var->data.location == VARYING_SLOT_PNTC ||
}
}
- if (c->s->stage == MESA_SHADER_VERTEX) {
+ if (c->s->info.stage == MESA_SHADER_VERTEX) {
assert(vpm_components_queued == 0);
assert(num_components == 0);
}
assert(array_len == 1);
(void)array_len;
- for (int i = 0; i < 4; i++)
- add_output(c, loc + i, var->data.location, i);
+ for (int i = 0; i < glsl_get_vector_elements(var->type); i++) {
+ add_output(c, loc + var->data.location_frac + i,
+ var->data.location,
+ var->data.location_frac + i);
+ }
- if (c->s->stage == MESA_SHADER_FRAGMENT) {
+ if (c->s->info.stage == MESA_SHADER_FRAGMENT) {
switch (var->data.location) {
case FRAG_RESULT_COLOR:
+ c->output_color_var[0] = var;
+ c->output_color_var[1] = var;
+ c->output_color_var[2] = var;
+ c->output_color_var[3] = var;
+ break;
case FRAG_RESULT_DATA0:
- c->output_color_var = var;
+ case FRAG_RESULT_DATA1:
+ case FRAG_RESULT_DATA2:
+ case FRAG_RESULT_DATA3:
+ c->output_color_var[var->data.location -
+ FRAG_RESULT_DATA0] = var;
break;
case FRAG_RESULT_DEPTH:
c->output_position_index = loc;
case nir_intrinsic_discard:
if (c->execute.file != QFILE_NULL) {
vir_PF(c, c->execute, V3D_QPU_PF_PUSHZ);
- vir_MOV_cond(c, V3D_QPU_COND_IFA, c->discard,
- vir_uniform_ui(c, ~0));
+ vir_set_cond(vir_SETMSF_dest(c, vir_reg(QFILE_NULL, 0),
+ vir_uniform_ui(c, 0)),
+ V3D_QPU_COND_IFA);
} else {
- vir_MOV_dest(c, c->discard, vir_uniform_ui(c, ~0));
+ vir_SETMSF_dest(c, vir_reg(QFILE_NULL, 0),
+ vir_uniform_ui(c, 0));
}
break;
*/
vir_PF(c, vir_AND(c, c->execute, vir_NOT(c, cond)),
V3D_QPU_PF_PUSHZ);
- vir_MOV_cond(c, V3D_QPU_COND_IFA, c->discard, cond);
+ vir_set_cond(vir_SETMSF_dest(c, vir_reg(QFILE_NULL, 0),
+ vir_uniform_ui(c, 0)),
+ V3D_QPU_COND_IFA);
} else {
- vir_OR_dest(c, c->discard, c->discard, cond);
+ vir_PF(c, cond, V3D_QPU_PF_PUSHZ);
+ vir_set_cond(vir_SETMSF_dest(c, vir_reg(QFILE_NULL, 0),
+ vir_uniform_ui(c, 0)),
+ V3D_QPU_COND_IFNA);
}
break;
static void
nir_to_vir(struct v3d_compile *c)
{
- if (c->s->stage == MESA_SHADER_FRAGMENT) {
+ if (c->s->info.stage == MESA_SHADER_FRAGMENT) {
c->payload_w = vir_MOV(c, vir_reg(QFILE_REG, 0));
c->payload_w_centroid = vir_MOV(c, vir_reg(QFILE_REG, 1));
c->payload_z = vir_MOV(c, vir_reg(QFILE_REG, 2));
- if (c->s->info.fs.uses_discard)
- c->discard = vir_MOV(c, vir_uniform_ui(c, 0));
-
if (c->fs_key->is_points) {
c->point_x = emit_fragment_varying(c, NULL, 0);
c->point_y = emit_fragment_varying(c, NULL, 0);
.lower_extract_word = true,
.lower_bitfield_insert = true,
.lower_bitfield_extract = true,
+ .lower_pack_unorm_2x16 = true,
+ .lower_pack_snorm_2x16 = true,
+ .lower_pack_unorm_4x8 = true,
+ .lower_pack_snorm_4x8 = true,
+ .lower_unpack_unorm_4x8 = true,
+ .lower_unpack_snorm_4x8 = true,
+ .lower_fdiv = true,
.lower_ffma = true,
.lower_flrp32 = true,
.lower_fpow = true,
.lower_fsat = true,
.lower_fsqrt = true,
- .lower_negate = true,
.native_integers = true,
};
v3d_nir_to_vir(struct v3d_compile *c)
{
if (V3D_DEBUG & (V3D_DEBUG_NIR |
- v3d_debug_flag_for_shader_stage(c->s->stage))) {
+ v3d_debug_flag_for_shader_stage(c->s->info.stage))) {
fprintf(stderr, "%s prog %d/%d NIR:\n",
vir_get_stage_name(c),
c->program_id, c->variant_id);
nir_to_vir(c);
- switch (c->s->stage) {
+ switch (c->s->info.stage) {
case MESA_SHADER_FRAGMENT:
emit_frag_end(c);
break;
}
if (V3D_DEBUG & (V3D_DEBUG_VIR |
- v3d_debug_flag_for_shader_stage(c->s->stage))) {
+ v3d_debug_flag_for_shader_stage(c->s->info.stage))) {
fprintf(stderr, "%s prog %d/%d pre-opt VIR:\n",
vir_get_stage_name(c),
c->program_id, c->variant_id);
/* XXX: vir_schedule_instructions(c); */
if (V3D_DEBUG & (V3D_DEBUG_VIR |
- v3d_debug_flag_for_shader_stage(c->s->stage))) {
+ v3d_debug_flag_for_shader_stage(c->s->info.stage))) {
fprintf(stderr, "%s prog %d/%d VIR:\n",
vir_get_stage_name(c),
c->program_id, c->variant_id);
projects / mesa.git / blobdiff