[Qemu-devel] [PATCH] ARM: fix carry flags for ARMv6 unsigned SIMD operations

[Vincent Palatin <vincent.palatin_qemu@polytechnique.org>]

[-- Attachment #1: Type: text/plain, Size: 521 bytes --]

Hi,

On ARMv6 emulation, I have caught some cases where the GE flags were
badly set after a "uadd8" operation.
After a quick code review, it seems to be a bad cut-n-paste between
16-bit and 8-bit UADD/USUB, indeed UADD8/USUB8 tries to set GE bits by
pair instead of one at a time.
Besides, the addition operations (UADD8/UADD16) set GE bits to "NOT
carry" instead of "carry" (probably once again due to a copy of the 
substraction code which sets flags to "NOT borrow")

I attach a patch to fix those issues.

--
Vincent

[-- Attachment #2: arith_ge.patch --]
[-- Type: text/plain, Size: 935 bytes --]

Index: target-arm/helper.c
===================================================================
--- target-arm/helper.c	(revision 4740)
+++ target-arm/helper.c	(working copy)
@@ -2059,7 +2059,7 @@
     uint32_t sum; \
     sum = (uint32_t)(uint16_t)(a) + (uint32_t)(uint16_t)(b); \
     RESULT(sum, n, 16); \
-    if ((sum >> 16) == 0) \
+    if ((sum >> 16) == 1) \
         ge |= 3 << (n * 2); \
     } while(0)
 
@@ -2067,8 +2067,8 @@
     uint32_t sum; \
     sum = (uint32_t)(uint8_t)(a) + (uint32_t)(uint8_t)(b); \
     RESULT(sum, n, 8); \
-    if ((sum >> 8) == 0) \
-        ge |= 3 << (n * 2); \
+    if ((sum >> 8) == 1) \
+        ge |= 1 << n; \
     } while(0)
 
 #define SUB16(a, b, n) do { \
@@ -2084,7 +2084,7 @@
     sum = (uint32_t)(uint8_t)(a) - (uint32_t)(uint8_t)(b); \
     RESULT(sum, n, 8); \
     if ((sum >> 8) == 0) \
-        ge |= 3 << (n * 2); \
+        ge |= 1 << n; \
     } while(0)
 
 #define PFX u