From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:51086) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1byYEs-000336-83 for qemu-devel@nongnu.org; Mon, 24 Oct 2016 01:56:03 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1byYEr-0004jF-DB for qemu-devel@nongnu.org; Mon, 24 Oct 2016 01:56:02 -0400 From: vijay.kilari@gmail.com Date: Mon, 24 Oct 2016 11:25:21 +0530 Message-Id: <1477288523-10819-2-git-send-email-vijay.kilari@gmail.com> In-Reply-To: <1477288523-10819-1-git-send-email-vijay.kilari@gmail.com> References: <1477288523-10819-1-git-send-email-vijay.kilari@gmail.com> Subject: [Qemu-devel] [PATCH v3 1/3] cutils: Set __builtin_prefetch optional parameters List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: qemu-arm@nongnu.org, peter.maydell@linaro.org, pbonzini@redhat.com, rth@twiddle.net Cc: qemu-devel@nongnu.org, vijay.kilari@gmail.com, Vijaya Kumar K From: Vijaya Kumar K Optional parameters of __builtin_prefetch() which specifies rw and locality to 0's. For checking buffer is zero, set rw as read and temporal locality to 0. On arm64, __builtin_prefetch(addr) generates 'prfm pldl1keep' where __builtin_prefetch(addr, 0, 0) generates 'prfm pldl1strm' instruction which is optimal for this use case Signed-off-by: Vijaya Kumar K --- util/bufferiszero.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/util/bufferiszero.c b/util/bufferiszero.c index eb974b7..421d945 100644 --- a/util/bufferiszero.c +++ b/util/bufferiszero.c @@ -49,7 +49,7 @@ buffer_zero_int(const void *buf, size_t len) const uint64_t *e = (uint64_t *)(((uintptr_t)buf + len) & -8); for (; p + 8 <= e; p += 8) { - __builtin_prefetch(p + 8); + __builtin_prefetch(p + 8, 0, 0); if (t) { return false; } @@ -86,7 +86,7 @@ buffer_zero_sse2(const void *buf, size_t len) /* Loop over 16-byte aligned blocks of 64. */ while (likely(p <= e)) { - __builtin_prefetch(p); + __builtin_prefetch(p, 0, 0); t = _mm_cmpeq_epi8(t, zero); if (unlikely(_mm_movemask_epi8(t) != 0xFFFF)) { return false; @@ -127,7 +127,7 @@ buffer_zero_sse4(const void *buf, size_t len) /* Loop over 16-byte aligned blocks of 64. */ while (likely(p <= e)) { - __builtin_prefetch(p); + __builtin_prefetch(p, 0, 0); if (unlikely(!_mm_testz_si128(t, t))) { return false; } @@ -162,7 +162,7 @@ buffer_zero_avx2(const void *buf, size_t len) if (likely(p <= e)) { /* Loop over 32-byte aligned blocks of 128. */ do { - __builtin_prefetch(p); + __builtin_prefetch(p, 0, 0); if (unlikely(!_mm256_testz_si256(t, t))) { return false; } @@ -303,7 +303,7 @@ bool buffer_is_zero(const void *buf, size_t len) } /* Fetch the beginning of the buffer while we select the accelerator. */ - __builtin_prefetch(buf); + __builtin_prefetch(buf, 0, 0); /* Use an optimized zero check if possible. Note that this also includes a check for an unrolled loop over 64-bit integers. */ -- 1.9.1