OpenHarmony/third_party_mesa3d
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

v3d: Use the core tex lowering.

Browse Source 
Even without any clever optimization on the unpack operations, this gives
us a useful value for the channels read field, which we can use to avoid
ldtmu instructions to the no-op register.

instructions in affected programs: 890712 -> 881974 (-0.98%)
This commit is contained in:
Eric Anholt
2018-12-20 09:43:43 -08:00
parent f217a94542
commit 8847370424
3 changed files with 10 additions and 123 deletions
Download Patch File Download Diff File Expand all files Collapse all files

68
src/broadcom/compiler/v3d40_tex.c
View File

@@ -166,18 +166,10 @@ v3d40_vir_emit_tex(struct v3d_compile *c, nir_tex_instr *instr)
/* Limit the number of channels returned to both how many the NIR
* instruction writes and how many the instruction could produce.
*
* XXX perf: Can we also limit to the number of channels that are
* actually read by the users of this NIR dest, so that we don't need
* to emit unused LDTMUs?
*/
uint32_t instr_return_channels = nir_tex_instr_dest_size(instr);
if (!p1_unpacked.output_type_32_bit)
instr_return_channels = (instr_return_channels + 1) / 2;
assert(instr->dest.is_ssa);
p0_unpacked.return_words_of_texture_data =
(1 << MIN2(instr_return_channels,
c->key->tex[unit].return_channels)) - 1;
nir_ssa_def_components_read(&instr->dest.ssa);
/* Word enables can't ask for more channels than the output type could
* provide (2 for f16, 4 for 32-bit).
@@ -232,62 +224,8 @@ v3d40_vir_emit_tex(struct v3d_compile *c, nir_tex_instr *instr)
while (tmu_writes > 16 / c->threads)
c->threads /= 2;
struct qreg return_values[4];
for (int i = 0; i < 4; i++) {
/* Swizzling .zw of an RG texture should give undefined
* results, not crash the compiler.
*/
if (p0_unpacked.return_words_of_texture_data & (1 << i))
return_values[i] = vir_LDTMU(c);
else
return_values[i] = c->undef;
}
for (int i = 0; i < nir_tex_instr_dest_size(instr); i++) {
struct qreg chan;
if (!p1_unpacked.output_type_32_bit) {
STATIC_ASSERT(PIPE_SWIZZLE_X == 0);
chan = return_values[i / 2];
/* XXX perf: We should move this unpacking into NIR.
* That would give us exposure of these types to NIR
* optimization, so that (for example) a repacking of
* half-float samples to the half-float render target
* could be eliminated.
*/
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);
} 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]);
}
ntq_store_dest(c, &instr->dest, i, chan);
ntq_store_dest(c, &instr->dest, i, vir_LDTMU(c));
}
}