File 0130-erts-Fix-spurious-exceptions-in-fnegate.patch of Package erlang

Overview Repositories Revisions Requests Users Attributes Meta

File 0130-erts-Fix-spurious-exceptions-in-fnegate.patch of Package erlang (Revision 57cf7c87cb80e2aaaf3f20327bf57529)

Currently displaying revision 57cf7c87cb80e2aaaf3f20327bf57529 , Show latest

From 3fbe8f43c2f4a5a48289ce06b4509468d75b8c0c Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?John=20H=C3=B6gberg?= <john@erlang.org>
Date: Fri, 10 Dec 2021 11:12:45 +0100
Subject: [PATCH] erts: Fix spurious exceptions in 'fnegate'

If we were to enter an 'fnegate' instruction with a non-finite
value in xmm0 by accident, we would erroneously throw an
exception.

I'm not certain whether this can occur in practice: 'fnegate' is
always preceded by another floating-point instruction that either
leaves a valid value in xmm0 or throws an exception (`fconv`,
`fload`, `fadd`, etc).

I can't think of a valid instruction sequence that would cause it,
but we'll fix it anyway just in case.
---
 erts/emulator/beam/jit/x86/instr_float.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/erts/emulator/beam/jit/x86/instr_float.cpp b/erts/emulator/beam/jit/x86/instr_float.cpp
index 43acf485ff..6302e3dcde 100644
--- a/erts/emulator/beam/jit/x86/instr_float.cpp
+++ b/erts/emulator/beam/jit/x86/instr_float.cpp
@@ -157,7 +157,7 @@ void BeamModuleAssembler::emit_i_fnegate(const ArgVal &Src, const ArgVal &Dst) {
     Label next = a.newLabel();
 
     /* xmm0 = 0.0 */
-    a.psubd(x86::xmm0, x86::xmm0);
+    a.pxor(x86::xmm0, x86::xmm0);
     a.movsd(x86::xmm1, getArgRef(Src));
     a.subpd(x86::xmm0, x86::xmm1);
 
-- 
2.31.1

Places

File 0130-erts-Fix-spurious-exceptions-in-fnegate.patch of Package erlang (Revision 57cf7c87cb80e2aaaf3f20327bf57529)

Places