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

i965/fs/nir: Simplify 64-bit store_output

Browse Source 
The swizzles weren't doing any good because swiz is just XYZW.  Also, we
were emitting an extra set of MOVs because shuffle_64bit_data_for_32bit
already does a MOV for us.  Finally, the temporary was only ever used
inside the inner loop so there's no need for it to actually be an array.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
This commit is contained in:
Jason Ekstrand
2017-08-26 11:26:40 -07:00
parent 18fde36ced
commit 6197a6b7ac
1 changed files with 6 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
src/intel/compiler/brw_fs_nir.cpp
View File

@@ -2568,7 +2568,6 @@ fs_visitor::nir_emit_tcs_intrinsic(const fs_builder &bld,
instr->src[0].ssa->bit_size : instr->src[0].reg.reg->bit_size) == 64; instr->src[0].ssa->bit_size : instr->src[0].reg.reg->bit_size) == 64;
fs_reg indirect_offset = get_indirect_offset(instr); fs_reg indirect_offset = get_indirect_offset(instr);
unsigned imm_offset = instr->const_index[0]; unsigned imm_offset = instr->const_index[0];
unsigned swiz = BRW_SWIZZLE_XYZW;
unsigned mask = instr->const_index[1]; unsigned mask = instr->const_index[1];
unsigned header_regs = 0; unsigned header_regs = 0;
fs_reg srcs[7]; fs_reg srcs[7];
@@ -2598,13 +2597,6 @@ fs_visitor::nir_emit_tcs_intrinsic(const fs_builder &bld,
} }
} }
/* 64-bit data needs to me shuffled before we can write it to the URB.
* We will use this temporary to shuffle the components in each
* iteration.
*/
fs_reg tmp =
fs_reg(VGRF, alloc.allocate(2 * iter_components), value.type);
mask = mask << first_component; mask = mask << first_component;
for (unsigned iter = 0; iter < num_iterations; iter++) { for (unsigned iter = 0; iter < num_iterations; iter++) {
@@ -2648,26 +2640,21 @@ fs_visitor::nir_emit_tcs_intrinsic(const fs_builder &bld,
continue; continue;
if (!is_64bit) { if (!is_64bit) {
srcs[header_regs + i + first_component] = srcs[header_regs + i + first_component] = offset(value, bld, i);
offset(value, bld, BRW_GET_SWZ(swiz, i));
} else { } else {
/* We need to shuffle the 64-bit data to match the layout /* We need to shuffle the 64-bit data to match the layout
* expected by our 32-bit URB write messages. We use a temporary * expected by our 32-bit URB write messages. We use a temporary
* for that. * for that.
*/ */
unsigned channel = BRW_GET_SWZ(swiz, iter * 2 + i); fs_reg dest = fs_reg(VGRF, alloc.allocate(2), value.type);
unsigned channel = iter * 2 + i;
shuffle_64bit_data_for_32bit_write(bld, shuffle_64bit_data_for_32bit_write(bld,
retype(offset(tmp, bld, 2 * i), BRW_REGISTER_TYPE_F), retype(dest, BRW_REGISTER_TYPE_F),
retype(offset(value, bld, 2 * channel), BRW_REGISTER_TYPE_DF), retype(offset(value, bld, 2 * channel), BRW_REGISTER_TYPE_DF),
1); 1);
/* Now copy the data to the destination */ srcs[header_regs + (i + first_component) * 2] = dest;
fs_reg dest = fs_reg(VGRF, alloc.allocate(2), value.type); srcs[header_regs + (i + first_component) * 2 + 1] =
unsigned idx = 2 * i;
bld.MOV(dest, offset(tmp, bld, idx));
bld.MOV(offset(dest, bld, 1), offset(tmp, bld, idx + 1));
srcs[header_regs + idx + first_component * 2] = dest;
srcs[header_regs + idx + 1 + first_component * 2] =
offset(dest, bld, 1); offset(dest, bld, 1);
} }
} }