From: Richard Henderson <richard.henderson@linaro.org>
To: BALATON Zoltan <balaton@eik.bme.hu>
Cc: qemu-devel@nongnu.org, alex.bennee@linaro.org, qemu-ppc@nongnu.org
Subject: Re: [PATCH] target/ppc: Merge COMPUTE_CLASS and COMPUTE_FPRF
Date: Tue, 23 May 2023 16:06:42 -0700 [thread overview]
Message-ID: <a05ba9d6-4344-d088-bfab-6c9f9ab3e0a4@linaro.org> (raw)
In-Reply-To: <e73e7d2d-275c-1489-1837-64c72db8b98d@eik.bme.hu>
On 5/23/23 16:02, BALATON Zoltan wrote:
> On Tue, 23 May 2023, Richard Henderson wrote:
>> Instead of computing an artifical "class" bitmask then
>> converting that to the fprf value, compute the final
>> value from the start.
>>
>> Reorder the tests to check the most likely cases first.
>>
>> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
>> ---
>> target/ppc/fpu_helper.c | 78 ++++++++++++-----------------------------
>> 1 file changed, 22 insertions(+), 56 deletions(-)
>>
>> diff --git a/target/ppc/fpu_helper.c b/target/ppc/fpu_helper.c
>> index a66e16c212..03150a0f10 100644
>> --- a/target/ppc/fpu_helper.c
>> +++ b/target/ppc/fpu_helper.c
>> @@ -141,62 +141,28 @@ static inline int ppc_float64_get_unbiased_exp(float64 f)
>> return ((f >> 52) & 0x7FF) - 1023;
>> }
>>
>> -/* Classify a floating-point number. */
>> -enum {
>> - is_normal = 1,
>> - is_zero = 2,
>> - is_denormal = 4,
>> - is_inf = 8,
>> - is_qnan = 16,
>> - is_snan = 32,
>> - is_neg = 64,
>> -};
>> -
>> -#define COMPUTE_CLASS(tp) \
>> -static int tp##_classify(tp arg) \
>> -{ \
>> - int ret = tp##_is_neg(arg) * is_neg; \
>> - if (unlikely(tp##_is_any_nan(arg))) { \
>> - float_status dummy = { }; /* snan_bit_is_one = 0 */ \
>> - ret |= (tp##_is_signaling_nan(arg, &dummy) \
>> - ? is_snan : is_qnan); \
>> - } else if (unlikely(tp##_is_infinity(arg))) { \
>> - ret |= is_inf; \
>> - } else if (tp##_is_zero(arg)) { \
>> - ret |= is_zero; \
>> - } else if (tp##_is_zero_or_denormal(arg)) { \
>> - ret |= is_denormal; \
>> - } else { \
>> - ret |= is_normal; \
>> - } \
>> - return ret; \
>> -}
>> -
>> -COMPUTE_CLASS(float16)
>> -COMPUTE_CLASS(float32)
>> -COMPUTE_CLASS(float64)
>> -COMPUTE_CLASS(float128)
>> -
>> -static void set_fprf_from_class(CPUPPCState *env, int class)
>> -{
>> - static const uint8_t fprf[6][2] = {
>> - { 0x04, 0x08 }, /* normalized */
>> - { 0x02, 0x12 }, /* zero */
>> - { 0x14, 0x18 }, /* denormalized */
>> - { 0x05, 0x09 }, /* infinity */
>> - { 0x11, 0x11 }, /* qnan */
>> - { 0x00, 0x00 }, /* snan -- flags are undefined */
>> - };
>> - bool isneg = class & is_neg;
>> -
>> - env->fpscr &= ~FP_FPRF;
>> - env->fpscr |= fprf[ctz32(class)][isneg] << FPSCR_FPRF;
>> -}
>> -
>> -#define COMPUTE_FPRF(tp) \
>> -void helper_compute_fprf_##tp(CPUPPCState *env, tp arg) \
>> -{ \
>> - set_fprf_from_class(env, tp##_classify(arg)); \
>> +#define COMPUTE_FPRF(tp) \
>> +void helper_compute_fprf_##tp(CPUPPCState *env, tp arg) \
>> +{ \
>> + bool neg = tp##_is_neg(arg); \
>> + target_ulong fprf; \
>> + if (likely(tp##_is_normal(arg))) { \
>> + fprf = neg ? 0x08 << FPSCR_FPRF : 0x04 << FPSCR_FPRF; \
>> + } else if (tp##_is_zero(arg)) { \
>> + fprf = neg ? 0x12 << FPSCR_FPRF : 0x02 << FPSCR_FPRF; \
>> + } else if (tp##_is_zero_or_denormal(arg)) { \
>> + fprf = neg ? 0x18 << FPSCR_FPRF : 0x14 << FPSCR_FPRF; \
>> + } else if (tp##_is_infinity(arg)) { \
>> + fprf = neg ? 0x09 << FPSCR_FPRF : 0x05 << FPSCR_FPRF; \
>> + } else { \
>> + float_status dummy = { }; /* snan_bit_is_one = 0 */ \
>> + if (tp##_is_signaling_nan(arg, &dummy)) { \
>> + fprf = 0x00 << FPSCR_FPRF; \
>> + } else { \
>> + fprf = 0x11 << FPSCR_FPRF; \
>
> If everything above is always shifted by FPSCR_FPRF then maybe it's easier to read with
> doing the shift once below an not in every case above.
I'm trying to make sure that the compiler generates all constants, instead of having a
runtime shift of a constant ...
>> + env->fpscr = (env->fpscr & ~FP_FPRF) | fprf; \
... here.
r~
next prev parent reply other threads:[~2023-05-23 23:07 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-05-23 20:25 [PATCH] target/ppc: Merge COMPUTE_CLASS and COMPUTE_FPRF Richard Henderson
2023-05-23 22:50 ` BALATON Zoltan
2023-05-23 23:02 ` BALATON Zoltan
2023-05-23 23:06 ` Richard Henderson [this message]
2023-05-24 8:52 ` BALATON Zoltan
2023-05-24 13:18 ` Alex Bennée
2023-05-28 10:05 ` Daniel Henrique Barboza
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=a05ba9d6-4344-d088-bfab-6c9f9ab3e0a4@linaro.org \
--to=richard.henderson@linaro.org \
--cc=alex.bennee@linaro.org \
--cc=balaton@eik.bme.hu \
--cc=qemu-devel@nongnu.org \
--cc=qemu-ppc@nongnu.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).