diff options
author | Bernhard Schommer <bernhardschommer@gmail.com> | 2022-02-07 15:10:54 +0100 |
---|---|---|
committer | Xavier Leroy <xavier.leroy@college-de-france.fr> | 2022-02-07 15:12:20 +0100 |
commit | 38b0425d524cd3e7260ac46e13153f007e8bc00d (patch) | |
tree | 987d9d3c42c6b0d7873115ec00fcdd99a714d7b5 /x86/SelectOpproof.v | |
parent | 87de3969856cf6a5eb5e1615ec817a015ed90d1c (diff) | |
download | compcert-kvx-38b0425d524cd3e7260ac46e13153f007e8bc00d.tar.gz compcert-kvx-38b0425d524cd3e7260ac46e13153f007e8bc00d.zip |
Add op for float max and min for x86.
The ops `Omaxf` and `Ominf` have the same semantics as `minsd` and
`maxsd` instruction, i.e. if both arguments are equal the second
argument is returned as well as for NaN.
The operations are the used in SelectOp to implement the built-in
function `__builtin_fmax` and `__builtin_fmin`.
Bug 32640
Diffstat (limited to 'x86/SelectOpproof.v')
-rw-r--r-- | x86/SelectOpproof.v | 7 |
1 files changed, 6 insertions, 1 deletions
diff --git a/x86/SelectOpproof.v b/x86/SelectOpproof.v index d8ab32a4..61cbf72b 100644 --- a/x86/SelectOpproof.v +++ b/x86/SelectOpproof.v @@ -1021,7 +1021,12 @@ Theorem eval_platform_builtin: platform_builtin_sem bf vl = Some v -> exists v', eval_expr ge sp e m le a v' /\ Val.lessdef v v'. Proof. - intros. discriminate. + intros until le. intros SEL ARGS SEM. + destruct bf; try discriminate. + - inv ARGS; try discriminate. inv H0; try discriminate. inv H2; try discriminate. + inv SEL. inv SEM. exists (minf v1 v0); split; auto. EvalOp. + - inv ARGS; try discriminate. inv H0; try discriminate. inv H2; try discriminate. + inv SEL. inv SEM. exists (maxf v1 v0); split; auto. EvalOp. Qed. End CMCONSTR. |