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

intel/fs: Refactor our shuffle emit code

Browse Source 
This adds an emit_scan_step helper which gives us a place to do
something a bit more interesting than emitting a single op.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7329>
This commit is contained in:
Jason Ekstrand
2020-10-26 11:58:32 -05:00
committed by Marge Bot
parent 44571c6a68
commit a6500236e3
1 changed files with 24 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files

57
src/intel/compiler/brw_fs_builder.h
View File

@@ -436,6 +436,18 @@ namespace brw {
return src_reg(dst); return src_reg(dst);
} }
void
emit_scan_step(enum opcode opcode, brw_conditional_mod mod,
const dst_reg &tmp,
unsigned left_offset, unsigned left_stride,
unsigned right_offset, unsigned right_stride) const
{
dst_reg left, right;
left = horiz_stride(horiz_offset(tmp, left_offset), left_stride);
right = horiz_stride(horiz_offset(tmp, right_offset), right_stride);
set_condmod(mod, emit(opcode, right, left, right));
}
void void
emit_scan(enum opcode opcode, const dst_reg &tmp, emit_scan(enum opcode opcode, const dst_reg &tmp,
unsigned cluster_size, brw_conditional_mod mod) const unsigned cluster_size, brw_conditional_mod mod) const
@@ -453,30 +465,23 @@ namespace brw {
ubld.emit_scan(opcode, left, cluster_size, mod); ubld.emit_scan(opcode, left, cluster_size, mod);
ubld.emit_scan(opcode, right, cluster_size, mod); ubld.emit_scan(opcode, right, cluster_size, mod);
if (cluster_size > half_width) { if (cluster_size > half_width) {
src_reg left_comp = component(left, half_width - 1); ubld.emit_scan_step(opcode, mod, tmp,
set_condmod(mod, ubld.emit(opcode, right, left_comp, right)); half_width - 1, 0, half_width, 1);
} }
return; return;
} }
if (cluster_size > 1) { if (cluster_size > 1) {
const fs_builder ubld = exec_all().group(dispatch_width() / 2, 0); const fs_builder ubld = exec_all().group(dispatch_width() / 2, 0);
const dst_reg left = horiz_stride(tmp, 2); ubld.emit_scan_step(opcode, mod, tmp, 0, 2, 1, 2);
const dst_reg right = horiz_stride(horiz_offset(tmp, 1), 2);
set_condmod(mod, ubld.emit(opcode, right, left, right));
} }
if (cluster_size > 2) { if (cluster_size > 2) {
if (type_sz(tmp.type) <= 4) { if (type_sz(tmp.type) <= 4) {
const fs_builder ubld = const fs_builder ubld =
exec_all().group(dispatch_width() / 4, 0); exec_all().group(dispatch_width() / 4, 0);
src_reg left = horiz_stride(horiz_offset(tmp, 1), 4); ubld.emit_scan_step(opcode, mod, tmp, 1, 4, 2, 4);
ubld.emit_scan_step(opcode, mod, tmp, 1, 4, 3, 4);
dst_reg right = horiz_stride(horiz_offset(tmp, 2), 4);
set_condmod(mod, ubld.emit(opcode, right, left, right));
right = horiz_stride(horiz_offset(tmp, 3), 4);
set_condmod(mod, ubld.emit(opcode, right, left, right));
} else { } else {
/* For 64-bit types, we have to do things differently because /* For 64-bit types, we have to do things differently because
* the code above would land us with destination strides that * the code above would land us with destination strides that
@@ -485,12 +490,8 @@ namespace brw {
* instructions. * instructions.
*/ */
const fs_builder ubld = exec_all().group(2, 0); const fs_builder ubld = exec_all().group(2, 0);
for (unsigned i = 0; i < dispatch_width(); i += 4)
for (unsigned i = 0; i < dispatch_width(); i += 4) { ubld.emit_scan_step(opcode, mod, tmp, i + 1, 0, i + 2, 1);
src_reg left = component(tmp, i + 1);
dst_reg right = horiz_offset(tmp, i + 2);
set_condmod(mod, ubld.emit(opcode, right, left, right));
}
} }
} }
@@ -498,24 +499,14 @@ namespace brw {
i < MIN2(cluster_size, dispatch_width()); i < MIN2(cluster_size, dispatch_width());
i *= 2) { i *= 2) {
const fs_builder ubld = exec_all().group(i, 0); const fs_builder ubld = exec_all().group(i, 0);
src_reg left = component(tmp, i - 1); ubld.emit_scan_step(opcode, mod, tmp, i - 1, 0, i, 1);
dst_reg right = horiz_offset(tmp, i);
set_condmod(mod, ubld.emit(opcode, right, left, right));
if (dispatch_width() > i * 2) { if (dispatch_width() > i * 2)
left = component(tmp, i * 3 - 1); ubld.emit_scan_step(opcode, mod, tmp, i * 3 - 1, 0, i * 3, 1);
right = horiz_offset(tmp, i * 3);
set_condmod(mod, ubld.emit(opcode, right, left, right));
}
if (dispatch_width() > i * 4) { if (dispatch_width() > i * 4) {
left = component(tmp, i * 5 - 1); ubld.emit_scan_step(opcode, mod, tmp, i * 5 - 1, 0, i * 5, 1);
right = horiz_offset(tmp, i * 5); ubld.emit_scan_step(opcode, mod, tmp, i * 7 - 1, 0, i * 7, 1);
set_condmod(mod, ubld.emit(opcode, right, left, right));
left = component(tmp, i * 7 - 1);
right = horiz_offset(tmp, i * 7);
set_condmod(mod, ubld.emit(opcode, right, left, right));
} }
} }
} }