fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-08 06:58:05 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

r600/sfn: make sure f2u32 is lowered late and correctly for 64 bit floats

Browse source 
With the latest changes in opt_algebraic we got f2u32 in the final code
that should be lowered before conversion to assembly.

Fixes: b3685f3ba7
    nir/algebraic: insert patterns inside optimizations list

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22640>
(cherry picked from commit 6a78af1dbb)
This commit is contained in:
Gert Wollny 2023-04-11 14:17:40 +02:00 • committed by Eric Engestrom
parent df6085eb9c
commit 4b6649c163
3 changed files with 6 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
.pick_status.json
View file

@ -121,7 +121,7 @@
"description": "r600/sfn: make sure f2u32 is lowered late and correctly for 64 bit floats",
"nominated": true,
"nomination_type": 1,
"resolution": 0,
"resolution": 1,
"main_sha": null,
"because_sha": "b3685f3ba7fddbe73f363ff4d53ca734841e4b06"
},

3
src/gallium/drivers/r600/sfn/sfn_nir.cpp
View file

@ -924,6 +924,9 @@ r600_shader_from_nir(struct r600_context *rctx,
while (optimize_once(sh))
;
if ((sh->info.bit_sizes_float | sh->info.bit_sizes_int) & 64)
NIR_PASS_V(sh, r600::r600_split_64bit_alu_and_phi);
bool late_algebraic_progress;
do {
late_algebraic_progress = false;

4
src/gallium/drivers/r600/sfn/sfn_nir_lower_64bit.cpp
View file

@ -212,11 +212,11 @@ class LowerSplit64op : public NirLowerInstruction {
* rounds, we have to remove the fractional part in the hi bits
* For values > UINT_MAX the result is undefined */
auto src = nir_ssa_for_alu_src(b, alu, 0);
src = nir_fsub(b, src, nir_ffract(b, src));
src = nir_fadd(b, src, nir_fneg(b, nir_ffract(b, src)));
auto gt0 = nir_flt(b, nir_imm_double(b, 0.0), src);
auto highval = nir_fmul_imm(b, src, 1.0 / 65536.0);
auto fract = nir_ffract(b, highval);
auto high = nir_f2u32(b, nir_f2f32(b, nir_fsub(b, highval, fract)));
auto high = nir_f2u32(b, nir_f2f32(b, nir_fadd(b, highval, nir_fneg(b, fract))));
auto lowval = nir_fmul_imm(b, fract, 65536.0);
auto low = nir_f2u32(b, nir_f2f32(b, lowval));
return nir_bcsel(b,