fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-05 07:28:11 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

r300: skip backend DCE for vertex shaders

Browse source 
Only mark the unused channels according to the writemask for now, but
this can go as well when we translate directly form NIR. Right now we
still need to do this separatelly as TGSI has no concept of unused
swizzle.

The old DCE pass was known to be buggy, so this also fixes at least
piglit glsl-vs-copy-propagation-1.

Reviewed-by: Filip Gawin <filip.gawin@collabora.com>
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/9279
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26848>
This commit is contained in:
Pavel Ondračka 2023-12-21 10:42:54 +01:00 • committed by Marge Bot
parent 0e2e4688af
commit 78404cc57d
4 changed files with 21 additions and 5 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

3
src/gallium/drivers/r300/ci/r300-rv530-nohiz-fails.txt
View file

@ -809,9 +809,6 @@ spec@glsl-1.10@execution@built-in-functions@vs-pow-vec4-vec4,Fail
spec@glsl-1.10@execution@clipping@clip-plane-transformation clipvert_pos,Fail
spec@glsl-1.10@execution@clipping@clip-plane-transformation pos_clipvert,Fail
# https://gitlab.freedesktop.org/mesa/mesa/-/issues/9279
spec@glsl-1.10@execution@copy-propagation@glsl-vs-copy-propagation-1,Crash
spec@glsl-1.10@execution@glsl-1.10-built-in-uniform-state,Fail
spec@glsl-1.10@execution@interpolation@interpolation-none-gl_backcolor-flat-vertex,Fail

2
src/gallium/drivers/r300/compiler/r3xx_vertprog.c
View file

@ -865,7 +865,7 @@ void r3xx_compile_vertex_program(struct r300_vertex_program_compiler *c)
{"add artificial outputs", 0, 1, rc_vs_add_artificial_outputs, NULL},
{"native rewrite", 1, 1, rc_local_transform, alu_rewrite},
{"emulate modifiers", 1, !is_r500, rc_local_transform, emulate_modifiers},
{"deadcode", 1, opt, rc_dataflow_deadcode, NULL},
{"unused channels", 1, opt, rc_mark_unused_channels, NULL},
{"dataflow optimize", 1, opt, rc_optimize, NULL},
/* This pass must be done after optimizations. */
{"source conflict resolve", 1, 1, rc_local_transform, resolve_src_conflicts},

19
src/gallium/drivers/r300/compiler/radeon_compiler.c
View file

@ -108,6 +108,25 @@ int rc_if_fail_helper(struct radeon_compiler * c, const char * file, int line, c
return 1;
}
void rc_mark_unused_channels(struct radeon_compiler * c, void *user)
{
unsigned int srcmasks[3];
for(struct rc_instruction * inst = c->Program.Instructions.Next;
inst != &c->Program.Instructions;
inst = inst->Next) {
rc_compute_sources_for_writemask(inst, inst->U.I.DstReg.WriteMask, srcmasks);
for(unsigned int src = 0; src < 3; ++src) {
for(unsigned int chan = 0; chan < 4; ++chan) {
if (!GET_BIT(srcmasks[src], chan))
SET_SWZ(inst->U.I.SrcReg[src].Swizzle, chan, RC_SWIZZLE_UNUSED);
}
}
}
}
/**
* Recompute c->Program.InputsRead and c->Program.OutputsWritten
* based on which inputs and outputs are actually referenced

2
src/gallium/drivers/r300/compiler/radeon_compiler.h
View file

@ -98,8 +98,8 @@ int rc_if_fail_helper(struct radeon_compiler * c, const char * file, int line, c
#define rc_assert(c, cond) \
(!(cond) && rc_if_fail_helper(c, __FILE__, __LINE__, #cond))
void rc_mark_unused_channels(struct radeon_compiler * c, void *user);
void rc_calculate_inputs_outputs(struct radeon_compiler * c);
void rc_copy_output(struct radeon_compiler * c, unsigned output, unsigned dup_output);
void rc_transform_fragment_wpos(struct radeon_compiler * c, unsigned wpos, unsigned new_input,
int full_vtransform);