From: David Hildenbrand <david@redhat.com>
To: Richard Henderson <richard.henderson@linaro.org>, qemu-devel@nongnu.org
Cc: qemu-s390x@nongnu.org, Cornelia Huck <cohuck@redhat.com>,
Thomas Huth <thuth@redhat.com>,
Richard Henderson <rth@twiddle.net>
Subject: Re: [Qemu-devel] [PATCH v1 2/5] s390x/tcg: Implement VECTOR FIND ELEMENT EQUAL
Date: Mon, 20 May 2019 11:17:40 +0200 [thread overview]
Message-ID: <066763ac-43c8-dea7-2762-c596bfc7c465@redhat.com> (raw)
In-Reply-To: <de30b861-b2ca-c83e-9058-1286047ab3d7@linaro.org>
On 17.05.19 19:42, Richard Henderson wrote:
> On 5/17/19 9:47 AM, Richard Henderson wrote:
>> first_equal = n;
>> first_zero = n;
>> for (i = n - 1; i >= 0; --i) {
>> if (data1 == data2) {
>> first_equal = i;
>> }
>> if (data1 == 0) {
>> first_zero = i;
>> }
>> }
>>
>> // As an aside, there are bit tricks for the above,
>> // but let's stay simple(r) for now.
>
> What the hell, it's not /that/ tricky.
>
>
> /*
> * Returns a bit set in the MSB of each element that is zero,
> * as defined by the mask M.
> */
> static inline uint64_t zero_search(uint64_t a, uint64_t m)
> {
> return ~(((a & m) + m) | a | m);
> }
>
> /*
> * Returns the byte offset for the first match, or 16 for no match.
> */
> static inline int match_index(uint64_t c0, uint64_t c1)
> {
> return (c0 ? clz64(c0) : clz64(c1) + 64) >> 3;
> }
>
> Use
>
> dup_const(MO_8, 0x7f)
> dup_const(MO_16, 0x7fff)
> dup_const(MO_32, 0x7fffffff)
>
> for the M parameter for the different element sizes.
>
> uint64_t a0, a1, b0, b1, e0, e1, z0, z1;
>
> a0 = s390_vec_read_element64(v2, 0);
> a1 = s390_vec_read_element64(v2, 1);
> b0 = s390_vec_read_element64(v3, 0);
> b1 = s390_vec_read_element64(v3, 1);
> e0 = zero_search(a0 ^ b0, m);
> e1 = zero_search(a1 ^ b1, m);
> first_equal = match_index(e0, e1);
>
> if (zs) {
> z0 = zero_search(a0, m);
> z1 = zero_search(a1, m);
> first_zero = match_index(z0, z1);
> ...
>
>
> r~
>
Crazy stuff, seems to work (not that I am surprised :D )
I now have:
+static int vfee(void *v1, const void *v2, const void *v3, bool zs,
uint8_t es)
+{
+ const uint64_t mask = dup_const(es, -1ull >> (65 - (1 << es) * 8));
+ uint64_t a0, a1, b0, b1, e0, e1, z0, z1;
+ uint64_t first_zero = 16;
+ uint64_t first_equal;
+
+ a0 = s390_vec_read_element64(v2, 0);
+ a1 = s390_vec_read_element64(v2, 1);
+ b0 = s390_vec_read_element64(v3, 0);
+ b1 = s390_vec_read_element64(v3, 1);
+ e0 = zero_search(a0 ^ b0, mask);
+ e1 = zero_search(a1 ^ b1, mask);
+ first_equal = match_index(e0, e1);
+
+ if (zs) {
+ z0 = zero_search(a0, mask);
+ z1 = zero_search(a1, mask);
+ first_zero = match_index(z0, z1);
+ }
+
+ /* zero out the destination vector */
+ s390_vec_write_element64(v1, 0, 0);
+ s390_vec_write_element64(v1, 1, 0);
+
+ if (first_zero == 16 && first_equal == 16) {
+ s390_vec_write_element8(v1, 7, 16);
+ return 3; /* no match */
+ } else if (first_zero == 16) {
+ s390_vec_write_element8(v1, 7, first_equal);
+ return 1; /* matching elements, no match for zero */
+ } else if (first_equal < first_zero) {
+ s390_vec_write_element8(v1, 7, first_equal);
+ return 2; /* matching elements before match for zero */
+ }
+ s390_vec_write_element8(v1, 7, first_zero);
+ return 0; /* match for zero */
+}
--
Thanks,
David / dhildenb
next prev parent reply other threads:[~2019-05-20 9:19 UTC|newest]
Thread overview: 27+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-05-15 20:31 [Qemu-devel] [PATCH v1 0/5] s390x/tcg: Vector Instruction Support Part 3 David Hildenbrand
2019-05-15 20:31 ` [Qemu-devel] [PATCH v1 1/5] s390x/tcg: Implement VECTOR FIND ANY ELEMENT EQUAL David Hildenbrand
2019-05-17 16:16 ` Richard Henderson
2019-05-20 9:51 ` David Hildenbrand
2019-05-22 11:01 ` David Hildenbrand
2019-05-22 11:09 ` Richard Henderson
2019-05-22 11:16 ` David Hildenbrand
2019-05-22 15:59 ` Richard Henderson
2019-05-22 18:16 ` David Hildenbrand
2019-05-22 18:46 ` Richard Henderson
2019-05-23 7:50 ` David Hildenbrand
2019-05-23 12:27 ` Richard Henderson
2019-05-23 12:34 ` David Hildenbrand
2019-05-23 12:59 ` David Hildenbrand
2019-05-23 13:50 ` Richard Henderson
2019-05-23 10:58 ` Alex Bennée
2019-05-15 20:31 ` [Qemu-devel] [PATCH v1 2/5] s390x/tcg: Implement VECTOR FIND " David Hildenbrand
2019-05-17 16:47 ` Richard Henderson
2019-05-17 17:42 ` Richard Henderson
2019-05-20 9:17 ` David Hildenbrand [this message]
2019-05-15 20:31 ` [Qemu-devel] [PATCH v1 3/5] s390x/tcg: Implement VECTOR FIND ELEMENT NOT EQUAL David Hildenbrand
2019-05-17 17:56 ` Richard Henderson
2019-05-20 9:48 ` David Hildenbrand
2019-05-15 20:31 ` [Qemu-devel] [PATCH v1 4/5] s390x/tcg: Implement VECTOR ISOLATE STRING David Hildenbrand
2019-05-17 18:20 ` Richard Henderson
2019-05-15 20:31 ` [Qemu-devel] [PATCH v1 5/5] s390x/tcg: Implement VECTOR STRING RANGE COMPARE David Hildenbrand
2019-05-17 18:37 ` Richard Henderson
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=066763ac-43c8-dea7-2762-c596bfc7c465@redhat.com \
--to=david@redhat.com \
--cc=cohuck@redhat.com \
--cc=qemu-devel@nongnu.org \
--cc=qemu-s390x@nongnu.org \
--cc=richard.henderson@linaro.org \
--cc=rth@twiddle.net \
--cc=thuth@redhat.com \
/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).