NeroReflex/mesa
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

nir: Fix cast

Browse source 
We were wrongly telling `nir_const_value_as_uint()` that `iter` had
`bit_size` bits, but in one case it is explicitly i64. This works on
little endian platforms, but caused the nir_loop_unroll_test.fadd{,_rev}
tests to fail on big endian platforms.

Bug: https://bugs.gentoo.org/921297
Fixes: 268ad47c11 ("nir/loop_analyze: Handle bit sizes correctly in calculate_iterations")
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26964>
This commit is contained in:
Matt Turner 2024-01-09 23:45:27 -05:00 • committed by Marge Bot
parent 3923d43908
commit 5997cf7587
1 changed files with 3 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

4
src/compiler/nir/nir_loop_analyze.c
View file

@ -827,6 +827,7 @@ get_iteration(nir_op cond_op, nir_const_value initial, nir_const_value step,
unsigned execution_mode)
{
nir_const_value span, iter;
unsigned iter_bit_size = bit_size;
switch (invert_comparison_if_needed(cond_op, invert_cond)) {
case nir_op_ine:
@ -880,13 +881,14 @@ get_iteration(nir_op cond_op, nir_const_value initial, nir_const_value step,
iter = eval_const_binop(nir_op_fdiv, bit_size, span,
step, execution_mode);
iter = eval_const_unop(nir_op_f2i64, bit_size, iter, execution_mode);
iter_bit_size = 64;
break;
default:
return -1;
}
uint64_t iter_u64 = nir_const_value_as_uint(iter, bit_size);
uint64_t iter_u64 = nir_const_value_as_uint(iter, iter_bit_size);
return iter_u64 > INT_MAX ? -1 : (int)iter_u64;
}