https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66731
Bug ID: 66731
Summary: vnmul, fnmul patterns incorrect for -frounding-math
Product: gcc
Version: 4.9.0
Status: UNCONFIRMED
Keywords: wrong-code
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: jsm28 at gcc dot gnu.org
Target Milestone: ---
Target: arm*-*-* aarch64*-*-*
config/arm/vfp.md has patterns:
(define_insn "*mulsf3negsf_vfp"
[(set (match_operand:SF 0 "s_register_operand" "=t")
(mult:SF (neg:SF (match_operand:SF 1 "s_register_operand" "t"))
(match_operand:SF 2 "s_register_operand" "t")))]
"TARGET_32BIT && TARGET_HARD_FLOAT && TARGET_VFP"
"vnmul%?.f32\\t%0, %1, %2"
[(set_attr "predicable" "yes")
(set_attr "predicable_short_it" "no")
(set_attr "type" "fmuls")]
)
(define_insn "*muldf3negdf_vfp"
[(set (match_operand:DF 0 "s_register_operand" "=w")
(mult:DF (neg:DF (match_operand:DF 1 "s_register_operand" "w"))
(match_operand:DF 2 "s_register_operand" "w")))]
"TARGET_32BIT && TARGET_HARD_FLOAT && TARGET_VFP_DOUBLE"
"vnmul%?.f64\\t%P0, %P1, %P2"
[(set_attr "predicable" "yes")
(set_attr "predicable_short_it" "no")
(set_attr "type" "fmuld")]
)
These describe the vnmul instruction as (-a) * b. It's actually -(a * b). The
two are different in round-upwards and round-downwards modes. This is
resulting in miscompilation of tgamma from current glibc git (which uses (-a) *
b, expecting that actual multiplication), and consequent test failures. On
inspection, AArch64 fnmul is similarly misdescribed, although I haven't done
any testing there.
