diff options
| author | Ian Romanick <[email protected]> | 2018-08-18 16:42:04 -0700 |
|---|---|---|
| committer | Ian Romanick <[email protected]> | 2019-05-06 22:52:29 -0700 |
| commit | d41cdef2a591930ad848b27f936aafcafc93b308 (patch) | |
| tree | c1527e05de52fc8c05b571c3dcfac7defb5c2d73 /src/gallium/drivers/vc4 | |
| parent | 158370ed2a012574bcc18e98075712da04fe22e8 (diff) | |
nir: Use the flrp lowering pass instead of nir_opt_algebraic
I tried to be very careful while updating all the various drivers, but I
don't have any of that hardware for testing. :(
i965 is the only platform that sets always_precise = true, and it is
only set true for fragment shaders. Gen4 and Gen5 both set lower_flrp32
only for vertex shaders. For fragment shaders, nir_op_flrp is lowered
during code generation as a(1-c)+bc. On all other platforms 64-bit
nir_op_flrp and on Gen11 32-bit nir_op_flrp are lowered using the old
nir_opt_algebraic method.
No changes on any other Intel platforms.
v2: Add panfrost changes.
Iron Lake and GM45 had similar results. (Iron Lake shown)
total cycles in shared programs: 188647754 -> 188647748 (<.01%)
cycles in affected programs: 5096 -> 5090 (-0.12%)
helped: 3
HURT: 0
helped stats (abs) min: 2 max: 2 x̄: 2.00 x̃: 2
helped stats (rel) min: 0.12% max: 0.12% x̄: 0.12% x̃: 0.12%
Reviewed-by: Matt Turner <[email protected]>
Diffstat (limited to 'src/gallium/drivers/vc4')
| -rw-r--r-- | src/gallium/drivers/vc4/vc4_program.c | 22 |
1 files changed, 22 insertions, 0 deletions
diff --git a/src/gallium/drivers/vc4/vc4_program.c b/src/gallium/drivers/vc4/vc4_program.c index 2ca3f907135..a2af55d6421 100644 --- a/src/gallium/drivers/vc4/vc4_program.c +++ b/src/gallium/drivers/vc4/vc4_program.c @@ -1527,6 +1527,10 @@ static void vc4_optimize_nir(struct nir_shader *s) { bool progress; + unsigned lower_flrp = + (s->options->lower_flrp16 ? 16 : 0) | + (s->options->lower_flrp32 ? 32 : 0) | + (s->options->lower_flrp64 ? 64 : 0); do { progress = false; @@ -1542,6 +1546,24 @@ vc4_optimize_nir(struct nir_shader *s) NIR_PASS(progress, s, nir_opt_peephole_select, 8, true, true); NIR_PASS(progress, s, nir_opt_algebraic); NIR_PASS(progress, s, nir_opt_constant_folding); + if (lower_flrp != 0) { + bool lower_flrp_progress; + + NIR_PASS(lower_flrp_progress, s, nir_lower_flrp, + lower_flrp, + false /* always_precise */, + s->options->lower_ffma); + if (lower_flrp_progress) { + NIR_PASS(progress, s, nir_opt_constant_folding); + progress = true; + } + + /* Nothing should rematerialize any flrps, so we only + * need to do this lowering once. + */ + lower_flrp = 0; + } + NIR_PASS(progress, s, nir_opt_undef); NIR_PASS(progress, s, nir_opt_loop_unroll, nir_var_shader_in | |