From: Richard Henderson <richard.henderson@linaro.org>
To: qemu-devel@nongnu.org
Cc: Alexander Monakov <amonakov@ispras.ru>,
Mikhail Romanov <mmromanov@ispras.ru>
Subject: [PATCH v6 02/10] util/bufferiszero: Remove AVX512 variant
Date: Wed, 24 Apr 2024 15:56:57 -0700 [thread overview]
Message-ID: <20240424225705.929812-3-richard.henderson@linaro.org> (raw)
In-Reply-To: <20240424225705.929812-1-richard.henderson@linaro.org>
From: Alexander Monakov <amonakov@ispras.ru>
Thanks to early checks in the inline buffer_is_zero wrapper, the SIMD
routines are invoked much more rarely in normal use when most buffers
are non-zero. This makes use of AVX512 unprofitable, as it incurs extra
frequency and voltage transition periods during which the CPU operates
at reduced performance, as described in
https://travisdowns.github.io/blog/2020/01/17/avxfreq1.html
Signed-off-by: Mikhail Romanov <mmromanov@ispras.ru>
Signed-off-by: Alexander Monakov <amonakov@ispras.ru>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20240206204809.9859-4-amonakov@ispras.ru>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
util/bufferiszero.c | 38 +++-----------------------------------
1 file changed, 3 insertions(+), 35 deletions(-)
diff --git a/util/bufferiszero.c b/util/bufferiszero.c
index f5a3634f9a..641d5f9b9e 100644
--- a/util/bufferiszero.c
+++ b/util/bufferiszero.c
@@ -64,7 +64,7 @@ buffer_zero_int(const void *buf, size_t len)
}
}
-#if defined(CONFIG_AVX512F_OPT) || defined(CONFIG_AVX2_OPT) || defined(__SSE2__)
+#if defined(CONFIG_AVX2_OPT) || defined(__SSE2__)
#include <immintrin.h>
/* Note that each of these vectorized functions require len >= 64. */
@@ -128,41 +128,12 @@ buffer_zero_avx2(const void *buf, size_t len)
}
#endif /* CONFIG_AVX2_OPT */
-#ifdef CONFIG_AVX512F_OPT
-static bool __attribute__((target("avx512f")))
-buffer_zero_avx512(const void *buf, size_t len)
-{
- /* Begin with an unaligned head of 64 bytes. */
- __m512i t = _mm512_loadu_si512(buf);
- __m512i *p = (__m512i *)(((uintptr_t)buf + 5 * 64) & -64);
- __m512i *e = (__m512i *)(((uintptr_t)buf + len) & -64);
-
- /* Loop over 64-byte aligned blocks of 256. */
- while (p <= e) {
- __builtin_prefetch(p);
- if (unlikely(_mm512_test_epi64_mask(t, t))) {
- return false;
- }
- t = p[-4] | p[-3] | p[-2] | p[-1];
- p += 4;
- }
-
- t |= _mm512_loadu_si512(buf + len - 4 * 64);
- t |= _mm512_loadu_si512(buf + len - 3 * 64);
- t |= _mm512_loadu_si512(buf + len - 2 * 64);
- t |= _mm512_loadu_si512(buf + len - 1 * 64);
-
- return !_mm512_test_epi64_mask(t, t);
-
-}
-#endif /* CONFIG_AVX512F_OPT */
-
/*
* Make sure that these variables are appropriately initialized when
* SSE2 is enabled on the compiler command-line, but the compiler is
* too old to support CONFIG_AVX2_OPT.
*/
-#if defined(CONFIG_AVX512F_OPT) || defined(CONFIG_AVX2_OPT)
+#if defined(CONFIG_AVX2_OPT)
# define INIT_USED 0
# define INIT_LENGTH 0
# define INIT_ACCEL buffer_zero_int
@@ -188,9 +159,6 @@ select_accel_cpuinfo(unsigned info)
unsigned len;
bool (*fn)(const void *, size_t);
} all[] = {
-#ifdef CONFIG_AVX512F_OPT
- { CPUINFO_AVX512F, 256, buffer_zero_avx512 },
-#endif
#ifdef CONFIG_AVX2_OPT
{ CPUINFO_AVX2, 128, buffer_zero_avx2 },
#endif
@@ -208,7 +176,7 @@ select_accel_cpuinfo(unsigned info)
return 0;
}
-#if defined(CONFIG_AVX512F_OPT) || defined(CONFIG_AVX2_OPT)
+#if defined(CONFIG_AVX2_OPT)
static void __attribute__((constructor)) init_accel(void)
{
used_accel = select_accel_cpuinfo(cpuinfo_init());
--
2.34.1
next prev parent reply other threads:[~2024-04-24 22:58 UTC|newest]
Thread overview: 21+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-04-24 22:56 [PATCH v6 00/10] Optimize buffer_is_zero Richard Henderson
2024-04-24 22:56 ` [PATCH v6 01/10] util/bufferiszero: Remove SSE4.1 variant Richard Henderson
2024-04-24 22:56 ` Richard Henderson [this message]
2024-04-29 11:16 ` [PATCH v6 02/10] util/bufferiszero: Remove AVX512 variant Daniel P. Berrangé
2024-04-29 11:29 ` Alexander Monakov
2024-04-24 22:56 ` [PATCH v6 03/10] util/bufferiszero: Reorganize for early test for acceleration Richard Henderson
2024-04-24 22:56 ` [PATCH v6 04/10] util/bufferiszero: Remove useless prefetches Richard Henderson
2024-04-24 22:57 ` [PATCH v6 05/10] util/bufferiszero: Optimize SSE2 and AVX2 variants Richard Henderson
2024-04-24 22:57 ` [PATCH v6 06/10] util/bufferiszero: Improve scalar variant Richard Henderson
2024-04-29 12:18 ` Philippe Mathieu-Daudé
2024-04-29 12:31 ` Richard Henderson
2024-04-29 13:21 ` Philippe Mathieu-Daudé
2024-04-24 22:57 ` [PATCH v6 07/10] util/bufferiszero: Introduce biz_accel_fn typedef Richard Henderson
2024-04-29 11:14 ` Philippe Mathieu-Daudé
2024-04-24 22:57 ` [PATCH v6 08/10] util/bufferiszero: Simplify test_buffer_is_zero_next_accel Richard Henderson
2024-04-29 11:05 ` Philippe Mathieu-Daudé
2024-04-24 22:57 ` [PATCH v6 09/10] util/bufferiszero: Add simd acceleration for aarch64 Richard Henderson
2024-04-29 11:06 ` Philippe Mathieu-Daudé
2024-04-29 12:45 ` Philippe Mathieu-Daudé
2024-04-24 22:57 ` [PATCH v6 10/10] tests/bench: Add bufferiszero-bench Richard Henderson
2024-04-29 11:12 ` Philippe Mathieu-Daudé
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=20240424225705.929812-3-richard.henderson@linaro.org \
--to=richard.henderson@linaro.org \
--cc=amonakov@ispras.ru \
--cc=mmromanov@ispras.ru \
--cc=qemu-devel@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).