On 2.4.19. 17:08, Philippe Mathieu-Daudé wrote:
Hi Mateja,
On 4/2/19 2:11 PM, Mateja Marjanovic wrote:
From: Mateja Marjanovic <mateja.marjano...@rt-rk.com>
MSA instructions DIV_<U|S>.<B|H|W|D> when dividing by zero,
didn't return the same value when executed on a referent hardware
(FPGA MIPS 64 r6, little endian) and when executed on QEMU, which
is not a real bug, because the result when dividing by zero is
UNPREDICTABLE [1] (page 141, 142).
I'm surprised by the arch, I'd have expected a MSA floating point
exception instead of UNPREDICTABLE.
Our architecture documents do not list any exceptions related to
division by zero, for instructions DIV_<U|S>.<B|H|W|D> and
MOD_<U|S>.<B|H|W|D>.
The only exceptions that are mentioned are:
Reserved Instruction Exception, MSA Disabled Exception. [1] (page 141,
142, 254, 255).
So here we decide to follow the FPGA model behavior rather than the
architecture...
If the community agree, my only request is to add a comment in the code
that this is the "FPGA MIPS 64 r6" behavior (else while looking at this
code later I'd be tempted to revert to a 0 return value instead of -1/1).
[1] MIPS Architecture for Programmers
Volume IV-j: The MIPS64 SIMD
Architecture Module, Revision 1.12
Signed-off-by: Mateja Marjanovic <mateja.marjano...@rt-rk.com>
---
target/mips/msa_helper.c | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/target/mips/msa_helper.c b/target/mips/msa_helper.c
index 655148d..8c77f12 100644
--- a/target/mips/msa_helper.c
+++ b/target/mips/msa_helper.c
@@ -641,14 +641,15 @@ static inline int64_t msa_div_s_df(uint32_t df, int64_t
arg1, int64_t arg2)
if (arg1 == DF_MIN_INT(df) && arg2 == -1) {
return DF_MIN_INT(df);
}
- return arg2 ? arg1 / arg2 : 0;
+ return arg2 ? arg1 / arg2
+ : arg1 >= 0 ? -1 : 1;
}
static inline int64_t msa_div_u_df(uint32_t df, int64_t arg1, int64_t arg2)
{
uint64_t u_arg1 = UNSIGNED(arg1, df);
uint64_t u_arg2 = UNSIGNED(arg2, df);
- return u_arg2 ? u_arg1 / u_arg2 : 0;
+ return arg2 ? u_arg1 / u_arg2 : -1;
}
static inline int64_t msa_mod_s_df(uint32_t df, int64_t arg1, int64_t arg2)
Thanks,
Mateja
