blob: 63d3090e9436016d2d7cd7db93760f10b7ccca00 (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
|
The Crunch cfnegs and cfnegd instructions have a hardware bug in all silicon
revisions (D0 to E2) whereby neg(0) returns 0 (not -0). See erratum 12.
For ieee-correctness, and to pass another case in GCC's IEEE testsuite,
we disable the instruction and do it in software unless
-funsafe-math-optimizations (included in -ffast-math) is given.
Martin Guy <martinwguy@yahoo.it>, September 2008
--- gcc-4.3.2/gcc/config/arm/cirrus.md-original 2007-06-12 17:01:24.000000000 +1000
+++ gcc-4.3.2/gcc/config/arm/cirrus.md 2007-06-12 17:03:26.000000000 +1000
@@ -254,18 +254,19 @@
[(set_attr "cirrus" "normal")]
)
+; Cirrus hardware bug: neg 0 -> 0 instead of -0
(define_insn "*cirrus_negsf2"
[(set (match_operand:SF 0 "cirrus_fp_register" "=v")
(neg:SF (match_operand:SF 1 "cirrus_fp_register" "v")))]
- "TARGET_32BIT && TARGET_HARD_FLOAT && TARGET_MAVERICK"
+ "TARGET_32BIT && TARGET_HARD_FLOAT && TARGET_MAVERICK && ! HONOR_SIGNED_ZEROS (SFmode)"
"cfnegs%?\\t%V0, %V1"
[(set_attr "cirrus" "normal")]
)
(define_insn "*cirrus_negdf2"
[(set (match_operand:DF 0 "cirrus_fp_register" "=v")
(neg:DF (match_operand:DF 1 "cirrus_fp_register" "v")))]
- "TARGET_32BIT && TARGET_HARD_FLOAT && TARGET_MAVERICK"
+ "TARGET_32BIT && TARGET_HARD_FLOAT && TARGET_MAVERICK && ! HONOR_SIGNED_ZEROS (DFmode)"
"cfnegd%?\\t%V0, %V1"
[(set_attr "cirrus" "normal")]
)
|
