Skip to contentSkip to conversation

Fix conversion from uint64 to float128

PATCH Applied Petr Tesarik 3 participants 4 messages May 11, 2018 – May 12, 2018

lore
Share & tools
lore

Petr Tesarik

author
#0 lore raw

PATCH 1/1 Fix conversion from uint64 to float128

The significand is passed to normalizeRoundAndPackFloat128() as high first, low second. The current code passes the integer first, so the result is incorrectly shifted left by 64 bits.

This bug affects the emulation of s390x instruction CXLGBR (convert from logical 64-bit binary-integer operand to extended BFP result).

Cc [email protected]
Signed-off-by Petr Tesarik <[email protected]>
---
 fpu/softfloat.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fpu/softfloat.c +1 -1
index 70e0c40a1c..386805475b 100644
@@ -3139,7 +3139,7 @@ float128 uint64_to_float128(uint64_t a, float_status *status)
     if (a == 0) {
         return float128_zero;
     }
-    return normalizeRoundAndPackFloat128(0, 0x406E, a, 0, status);
+    return normalizeRoundAndPackFloat128(0, 0x406E, 0, a, status);
 }
Signature · 1 line
2.13.6
author
#1 lore raw
Quoted 6 lines · On Fri, 11 May 2018 09:10:52 +0200 Petr Tesarik <[email protected]> wrote:

The significand is passed to normalizeRoundAndPackFloat128() as high first, low second. The current code passes the integer first, so the result is incorrectly shifted left by 64 bits.

This bug affects the emulation of s390x instruction CXLGBR (convert from logical 64-bit binary-integer operand to extended BFP result).

I forgot to add a simple reproducer:

#include <stdio.h>
#include <stdlib.h>

int main(int argc, char **argv)

{
        unsigned long long x = atoll(argv[1]);
        long double d = x;

        printf("%llu -> %Lf\n", x, d);

        return 0;
}

On a real s390x, I get this:
linux-nig1:~ # ./cast 1
1 -> 1.000000

An s390x emulated with (unpatched) qemu-system-s390x gives:
linux-2s2c:~ # ./cast 1
1 -> 18446744073709551616.000000

That is 2^64.

Petr T

Reviewed-by on v1 1/1
#2 lore raw
In reply to Petr Tesarik
Quoted 18 lines · Petr Tesarik

The significand is passed to normalizeRoundAndPackFloat128() as high first, low second. The current code passes the integer first, so the result is incorrectly shifted left by 64 bits.

This bug affects the emulation of s390x instruction CXLGBR (convert from logical 64-bit binary-integer operand to extended BFP result).

Cc [email protected]
Signed-off-by Petr Tesarik <[email protected]>
---
 fpu/softfloat.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fpu/softfloat.c +0 -0
index 70e0c40a1c..386805475b 100644

fpu/softfloat.c float128 uint64_to_float128(uint64_t a, float_status *status) · line 3143

Quoted 6 lines
fpu/softfloat.c +1 -1
@@ -3139,7 +3139,7 @@ float128 uint64_to_float128(uint64_t a, float_status *status)
     if (a == 0) {
         return float128_zero;
     }
-    return normalizeRoundAndPackFloat128(0, 0x406E, a, 0, status);
+    return normalizeRoundAndPackFloat128(0, 0x406E, 0, a, status);
 }
Reviewed-by Peter Maydell <[email protected]>

thanks
-- PMM

Applied
#3 lore raw
In reply to Petr Tesarik
Quoted 4 lines · Petr Tesarik

The significand is passed to normalizeRoundAndPackFloat128() as high first, low second. The current code passes the integer first, so the result is incorrectly shifted left by 64 bits.

Quoted 8 lines

This bug affects the emulation of s390x instruction CXLGBR (convert from logical 64-bit binary-integer operand to extended BFP result).

Cc [email protected]
Signed-off-by Petr Tesarik <[email protected]>
---
 fpu/softfloat.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Queued, thanks.

r~