x86: 0.855
architecture: 0.770
mistranslation: 0.767
i386: 0.747
device: 0.640
kernel: 0.567
user-level: 0.548
graphic: 0.544
hypervisor: 0.539
socket: 0.539
ppc: 0.535
files: 0.525
performance: 0.524
peripherals: 0.507
vnc: 0.495
semantic: 0.495
risc-v: 0.448
PID: 0.429
network: 0.411
permissions: 0.383
arm: 0.382
VMM: 0.359
TCG: 0.331
boot: 0.317
register: 0.316
debug: 0.312
virtual: 0.310
KVM: 0.296
assembly: 0.277

x86: FPU_MAX, FPU_MIN incorrect

Dear All,

Bug was found in qemu.git.
Now (0.15, 1.0) all fpu is softfpu.
See target-i386/ops_sse.h:
#define FPU_MIN(size, a, b) (a) < (b) ? (a) : (b)
#define FPU_MAX(size, a, b) (a) > (b) ? (a) : (b)
It is incorrect now, becouse float64 (or 32) is (typedef) uint64_t (or 32).
And if we have signed operands we get error...

There is a test with this error (spec shinx3 test data, results diffs on machine and qemu (linux)) and fixed patch. See attach.

Daniil.





misprint:
spec sphinx3 test data

482.sphinx3: http://www.spec.org/cpu2006/CFP2006/

The attached patch is incorrect (using the softfloat _min/_max functions will give wrong answers for some special cases). The correct macros are
#define FPU_MIN(size, a, b) float ## size ## _lt(a, b, &env->sse_status) ? (a) : (b)
#define FPU_MAX(size, a, b) float ## size ## _lt(b, a, &env->sse_status) ? (a) : (b)

(see recent discussion on the qemu-devel list).


Hello!
Can I commit this patch (in development branch), and close this bug...
Or you must do it?

Your patch is broken, as I said before.


Yes, but you patch is correct...

May be commit you patch

I say about
#define FPU_MIN(size, a, b) float ## size ## _lt(a, b, &env->sse_status) ? (a) : (b)
#define FPU_MAX(size, a, b) float ## size ## _lt(b, a, &env->sse_status) ? (b) : (a)

The patch mentioned in comment #5 has been included here:
http://git.qemu.org/?p=qemu.git;a=commitdiff;h=a4d1f142542935b90d2e