From 43a123c1ae6613b3efeed291fa552ecd909d3acf Mon Sep 17 00:00:00 2001
From: Daniel Baumann <daniel.baumann@progress-linux.org>
Date: Tue, 16 Apr 2024 21:23:18 +0200
Subject: Adding upstream version 1.20.14.

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
---
 src/hash/crc32/crc32_s390x.go | 91 +++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 91 insertions(+)
 create mode 100644 src/hash/crc32/crc32_s390x.go

(limited to 'src/hash/crc32/crc32_s390x.go')

diff --git a/src/hash/crc32/crc32_s390x.go b/src/hash/crc32/crc32_s390x.go
new file mode 100644
index 0000000..4e50b56
--- /dev/null
+++ b/src/hash/crc32/crc32_s390x.go
@@ -0,0 +1,91 @@
+// Copyright 2016 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package crc32
+
+import "internal/cpu"
+
+const (
+	vxMinLen    = 64
+	vxAlignMask = 15 // align to 16 bytes
+)
+
+// hasVX reports whether the machine has the z/Architecture
+// vector facility installed and enabled.
+var hasVX = cpu.S390X.HasVX
+
+// vectorizedCastagnoli implements CRC32 using vector instructions.
+// It is defined in crc32_s390x.s.
+//
+//go:noescape
+func vectorizedCastagnoli(crc uint32, p []byte) uint32
+
+// vectorizedIEEE implements CRC32 using vector instructions.
+// It is defined in crc32_s390x.s.
+//
+//go:noescape
+func vectorizedIEEE(crc uint32, p []byte) uint32
+
+func archAvailableCastagnoli() bool {
+	return hasVX
+}
+
+var archCastagnoliTable8 *slicing8Table
+
+func archInitCastagnoli() {
+	if !hasVX {
+		panic("not available")
+	}
+	// We still use slicing-by-8 for small buffers.
+	archCastagnoliTable8 = slicingMakeTable(Castagnoli)
+}
+
+// archUpdateCastagnoli calculates the checksum of p using
+// vectorizedCastagnoli.
+func archUpdateCastagnoli(crc uint32, p []byte) uint32 {
+	if !hasVX {
+		panic("not available")
+	}
+	// Use vectorized function if data length is above threshold.
+	if len(p) >= vxMinLen {
+		aligned := len(p) & ^vxAlignMask
+		crc = vectorizedCastagnoli(crc, p[:aligned])
+		p = p[aligned:]
+	}
+	if len(p) == 0 {
+		return crc
+	}
+	return slicingUpdate(crc, archCastagnoliTable8, p)
+}
+
+func archAvailableIEEE() bool {
+	return hasVX
+}
+
+var archIeeeTable8 *slicing8Table
+
+func archInitIEEE() {
+	if !hasVX {
+		panic("not available")
+	}
+	// We still use slicing-by-8 for small buffers.
+	archIeeeTable8 = slicingMakeTable(IEEE)
+}
+
+// archUpdateIEEE calculates the checksum of p using vectorizedIEEE.
+func archUpdateIEEE(crc uint32, p []byte) uint32 {
+	if !hasVX {
+		panic("not available")
+	}
+	// Use vectorized function if data length is above threshold.
+	if len(p) >= vxMinLen {
+		aligned := len(p) & ^vxAlignMask
+		crc = vectorizedIEEE(crc, p[:aligned])
+		p = p[aligned:]
+	}
+	if len(p) == 0 {
+		return crc
+	}
+	return slicingUpdate(crc, archIeeeTable8, p)
+}
-- 
cgit v1.2.3