diff options
| author | Ian Romanick <ian.d.romanick@intel.com> | 2022-10-12 14:21:02 -0700 |
|---|---|---|
| committer | Marge Bot <emma+marge@anholt.net> | 2023-03-10 15:27:17 +0000 |
| commit | 0cadc3830fa617750d96c133391cc82ff26d531a (patch) | |
| tree | 1ae24d03b36a8d576ef4164382680c15b530d634 /src/compiler/nir | |
| parent | 831f9d3f615ca5228e9614191d76f95735bfc2cc (diff) | |
nir/lower_int64: Optionally lower ufind_msb using uadd_sat
v2: Fix inverted condition for applying the optimization. Noticed by
Ken.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19042>
Diffstat (limited to 'src/compiler/nir')
| -rw-r--r-- | src/compiler/nir/nir_lower_int64.c | 22 |
1 files changed, 19 insertions, 3 deletions
diff --git a/src/compiler/nir/nir_lower_int64.c b/src/compiler/nir/nir_lower_int64.c index 422bd40e566..eb1dc1e3409 100644 --- a/src/compiler/nir/nir_lower_int64.c +++ b/src/compiler/nir/nir_lower_int64.c @@ -680,9 +680,25 @@ lower_ufind_msb64(nir_builder *b, nir_ssa_def *x) nir_ssa_def *x_hi = nir_unpack_64_2x32_split_y(b, x); nir_ssa_def *lo_count = nir_ufind_msb(b, x_lo); nir_ssa_def *hi_count = nir_ufind_msb(b, x_hi); - nir_ssa_def *valid_hi_bits = nir_ine(b, x_hi, nir_imm_int(b, 0)); - nir_ssa_def *hi_res = nir_iadd(b, nir_imm_intN_t(b, 32, 32), hi_count); - return nir_bcsel(b, valid_hi_bits, hi_res, lo_count); + + if (b->shader->options->lower_uadd_sat) { + nir_ssa_def *valid_hi_bits = nir_ine(b, x_hi, nir_imm_int(b, 0)); + nir_ssa_def *hi_res = nir_iadd(b, nir_imm_intN_t(b, 32, 32), hi_count); + return nir_bcsel(b, valid_hi_bits, hi_res, lo_count); + } else { + /* If hi_count was -1, it will still be -1 after this uadd_sat. As a + * result, hi_count is either -1 or the correct return value for 64-bit + * ufind_msb. + */ + nir_ssa_def *hi_res = nir_uadd_sat(b, nir_imm_intN_t(b, 32, 32), hi_count); + + /* hi_res is either -1 or a value in the range [63, 32]. lo_count is + * either -1 or a value in the range [31, 0]. The imax will pick + * lo_count only when hi_res is -1. In those cases, lo_count is + * guaranteed to be the correct answer. + */ + return nir_imax(b, hi_res, lo_count); + } } static nir_ssa_def * |