lib/raid6: Add AVX2 optimized gen_syndrome functions

Add AVX2 optimized gen_syndrom functions, which is simply based on
sse2.c written by hpa.

Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Jim Kukunas <james.t.kukunas@linux.intel.com>
Signed-off-by: NeilBrown <neilb@suse.de>

4 files changed, 272 insertions, 2 deletions

diff --git a/lib/raid6/Makefile b/lib/raid6/Makefile
index 8c2e22b..3430711 100644
--- a/lib/raid6/Makefile
+++ b/lib/raid6/Makefile
@@ -2,7 +2,7 @@ obj-$(CONFIG_RAID6_PQ)	+= raid6_pq.o
 
 raid6_pq-y	+= algos.o recov.o recov_ssse3.o recov_avx2.o tables.o int1.o int2.o int4.o \
 		   int8.o int16.o int32.o altivec1.o altivec2.o altivec4.o \
-		   altivec8.o mmx.o sse1.o sse2.o
+		   altivec8.o mmx.o sse1.o sse2.o avx2.o
 hostprogs-y	+= mktables
 
 quiet_cmd_unroll = UNROLL  $@
diff --git a/lib/raid6/algos.c b/lib/raid6/algos.c
index 8b7f55c..6d7316f 100644
--- a/lib/raid6/algos.c
+++ b/lib/raid6/algos.c
@@ -45,11 +45,20 @@ const struct raid6_calls * const raid6_algos[] = {
 	&raid6_sse1x2,
 	&raid6_sse2x1,
 	&raid6_sse2x2,
+#ifdef CONFIG_AS_AVX2
+	&raid6_avx2x1,
+	&raid6_avx2x2,
+#endif
 #endif
 #if defined(__x86_64__) && !defined(__arch_um__)
 	&raid6_sse2x1,
 	&raid6_sse2x2,
 	&raid6_sse2x4,
+#ifdef CONFIG_AS_AVX2
+	&raid6_avx2x1,
+	&raid6_avx2x2,
+	&raid6_avx2x4,
+#endif
 #endif
 #ifdef CONFIG_ALTIVEC
 	&raid6_altivec1,
diff --git a/lib/raid6/avx2.c b/lib/raid6/avx2.c
new file mode 100644
index 0000000..bc3b1dd
--- /dev/null
+++ b/lib/raid6/avx2.c
@@ -0,0 +1,251 @@
+/* -*- linux-c -*- ------------------------------------------------------- *
+ *
+ *   Copyright (C) 2012 Intel Corporation
+ *   Author: Yuanhan Liu <yuanhan.liu@linux.intel.com>
+ *
+ *   Based on sse2.c: Copyright 2002 H. Peter Anvin - All Rights Reserved
+ *
+ *
+ *   This program is free software; you can redistribute it and/or modify
+ *   it under the terms of the GNU General Public License as published by
+ *   the Free Software Foundation, Inc., 53 Temple Place Ste 330,
+ *   Boston MA 02111-1307, USA; either version 2 of the License, or
+ *   (at your option) any later version; incorporated herein by reference.
+ *
+ * ----------------------------------------------------------------------- */
+
+/*
+ * AVX2 implementation of RAID-6 syndrome functions
+ *
+ */
+
+#ifdef CONFIG_AS_AVX2
+
+#include <linux/raid/pq.h>
+#include "x86.h"
+
+static const struct raid6_avx2_constants {
+	u64 x1d[4];
+} raid6_avx2_constants __aligned(32) = {
+	{ 0x1d1d1d1d1d1d1d1dULL, 0x1d1d1d1d1d1d1d1dULL,
+	  0x1d1d1d1d1d1d1d1dULL, 0x1d1d1d1d1d1d1d1dULL,},
+};
+
+static int raid6_have_avx2(void)
+{
+	return boot_cpu_has(X86_FEATURE_AVX2) && boot_cpu_has(X86_FEATURE_AVX);
+}
+
+/*
+ * Plain AVX2 implementation
+ */
+static void raid6_avx21_gen_syndrome(int disks, size_t bytes, void **ptrs)
+{
+	u8 **dptr = (u8 **)ptrs;
+	u8 *p, *q;
+	int d, z, z0;
+
+	z0 = disks - 3;		/* Highest data disk */
+	p = dptr[z0+1];		/* XOR parity */
+	q = dptr[z0+2];		/* RS syndrome */
+
+	kernel_fpu_begin();
+
+	asm volatile("vmovdqa %0,%%ymm0" : : "m" (raid6_avx2_constants.x1d[0]));
+	asm volatile("vpxor %ymm3,%ymm3,%ymm3");	/* Zero temp */
+
+	for (d = 0; d < bytes; d += 32) {
+		asm volatile("prefetchnta %0" : : "m" (dptr[z0][d]));
+		asm volatile("vmovdqa %0,%%ymm2" : : "m" (dptr[z0][d]));/* P[0] */
+		asm volatile("prefetchnta %0" : : "m" (dptr[z0-1][d]));
+		asm volatile("vmovdqa %ymm2,%ymm4");/* Q[0] */
+		asm volatile("vmovdqa %0,%%ymm6" : : "m" (dptr[z0-1][d]));
+		for (z = z0-2; z >= 0; z--) {
+			asm volatile("prefetchnta %0" : : "m" (dptr[z][d]));
+			asm volatile("vpcmpgtb %ymm4,%ymm3,%ymm5");
+			asm volatile("vpaddb %ymm4,%ymm4,%ymm4");
+			asm volatile("vpand %ymm0,%ymm5,%ymm5");
+			asm volatile("vpxor %ymm5,%ymm4,%ymm4");
+			asm volatile("vpxor %ymm6,%ymm2,%ymm2");
+			asm volatile("vpxor %ymm6,%ymm4,%ymm4");
+			asm volatile("vmovdqa %0,%%ymm6" : : "m" (dptr[z][d]));
+		}
+		asm volatile("vpcmpgtb %ymm4,%ymm3,%ymm5");
+		asm volatile("vpaddb %ymm4,%ymm4,%ymm4");
+		asm volatile("vpand %ymm0,%ymm5,%ymm5");
+		asm volatile("vpxor %ymm5,%ymm4,%ymm4");
+		asm volatile("vpxor %ymm6,%ymm2,%ymm2");
+		asm volatile("vpxor %ymm6,%ymm4,%ymm4");
+
+		asm volatile("vmovntdq %%ymm2,%0" : "=m" (p[d]));
+		asm volatile("vpxor %ymm2,%ymm2,%ymm2");
+		asm volatile("vmovntdq %%ymm4,%0" : "=m" (q[d]));
+		asm volatile("vpxor %ymm4,%ymm4,%ymm4");
+	}
+
+	asm volatile("sfence" : : : "memory");
+	kernel_fpu_end();
+}
+
+const struct raid6_calls raid6_avx2x1 = {
+	raid6_avx21_gen_syndrome,
+	raid6_have_avx2,
+	"avx2x1",
+	1			/* Has cache hints */
+};
+
+/*
+ * Unrolled-by-2 AVX2 implementation
+ */
+static void raid6_avx22_gen_syndrome(int disks, size_t bytes, void **ptrs)
+{
+	u8 **dptr = (u8 **)ptrs;
+	u8 *p, *q;
+	int d, z, z0;
+
+	z0 = disks - 3;		/* Highest data disk */
+	p = dptr[z0+1];		/* XOR parity */
+	q = dptr[z0+2];		/* RS syndrome */
+
+	kernel_fpu_begin();
+
+	asm volatile("vmovdqa %0,%%ymm0" : : "m" (raid6_avx2_constants.x1d[0]));
+	asm volatile("vpxor %ymm1,%ymm1,%ymm1"); /* Zero temp */
+
+	/* We uniformly assume a single prefetch covers at least 32 bytes */
+	for (d = 0; d < bytes; d += 64) {
+		asm volatile("prefetchnta %0" : : "m" (dptr[z0][d]));
+		asm volatile("prefetchnta %0" : : "m" (dptr[z0][d+32]));
+		asm volatile("vmovdqa %0,%%ymm2" : : "m" (dptr[z0][d]));/* P[0] */
+		asm volatile("vmovdqa %0,%%ymm3" : : "m" (dptr[z0][d+32]));/* P[1] */
+		asm volatile("vmovdqa %ymm2,%ymm4"); /* Q[0] */
+		asm volatile("vmovdqa %ymm3,%ymm6"); /* Q[1] */
+		for (z = z0-1; z >= 0; z--) {
+			asm volatile("prefetchnta %0" : : "m" (dptr[z][d]));
+			asm volatile("prefetchnta %0" : : "m" (dptr[z][d+32]));
+			asm volatile("vpcmpgtb %ymm4,%ymm1,%ymm5");
+			asm volatile("vpcmpgtb %ymm6,%ymm1,%ymm7");
+			asm volatile("vpaddb %ymm4,%ymm4,%ymm4");
+			asm volatile("vpaddb %ymm6,%ymm6,%ymm6");
+			asm volatile("vpand %ymm0,%ymm5,%ymm5");
+			asm volatile("vpand %ymm0,%ymm7,%ymm7");
+			asm volatile("vpxor %ymm5,%ymm4,%ymm4");
+			asm volatile("vpxor %ymm7,%ymm6,%ymm6");
+			asm volatile("vmovdqa %0,%%ymm5" : : "m" (dptr[z][d]));
+			asm volatile("vmovdqa %0,%%ymm7" : : "m" (dptr[z][d+32]));
+			asm volatile("vpxor %ymm5,%ymm2,%ymm2");
+			asm volatile("vpxor %ymm7,%ymm3,%ymm3");
+			asm volatile("vpxor %ymm5,%ymm4,%ymm4");
+			asm volatile("vpxor %ymm7,%ymm6,%ymm6");
+		}
+		asm volatile("vmovntdq %%ymm2,%0" : "=m" (p[d]));
+		asm volatile("vmovntdq %%ymm3,%0" : "=m" (p[d+32]));
+		asm volatile("vmovntdq %%ymm4,%0" : "=m" (q[d]));
+		asm volatile("vmovntdq %%ymm6,%0" : "=m" (q[d+32]));
+	}
+
+	asm volatile("sfence" : : : "memory");
+	kernel_fpu_end();
+}
+
+const struct raid6_calls raid6_avx2x2 = {
+	raid6_avx22_gen_syndrome,
+	raid6_have_avx2,
+	"avx2x2",
+	1			/* Has cache hints */
+};
+
+#ifdef CONFIG_X86_64
+
+/*
+ * Unrolled-by-4 AVX2 implementation
+ */
+static void raid6_avx24_gen_syndrome(int disks, size_t bytes, void **ptrs)
+{
+	u8 **dptr = (u8 **)ptrs;
+	u8 *p, *q;
+	int d, z, z0;
+
+	z0 = disks - 3;		/* Highest data disk */
+	p = dptr[z0+1];		/* XOR parity */
+	q = dptr[z0+2];		/* RS syndrome */
+
+	kernel_fpu_begin();
+
+	asm volatile("vmovdqa %0,%%ymm0" : : "m" (raid6_avx2_constants.x1d[0]));
+	asm volatile("vpxor %ymm1,%ymm1,%ymm1");	/* Zero temp */
+	asm volatile("vpxor %ymm2,%ymm2,%ymm2");	/* P[0] */
+	asm volatile("vpxor %ymm3,%ymm3,%ymm3");	/* P[1] */
+	asm volatile("vpxor %ymm4,%ymm4,%ymm4");	/* Q[0] */
+	asm volatile("vpxor %ymm6,%ymm6,%ymm6");	/* Q[1] */
+	asm volatile("vpxor %ymm10,%ymm10,%ymm10");	/* P[2] */
+	asm volatile("vpxor %ymm11,%ymm11,%ymm11");	/* P[3] */
+	asm volatile("vpxor %ymm12,%ymm12,%ymm12");	/* Q[2] */
+	asm volatile("vpxor %ymm14,%ymm14,%ymm14");	/* Q[3] */
+
+	for (d = 0; d < bytes; d += 128) {
+		for (z = z0; z >= 0; z--) {
+			asm volatile("prefetchnta %0" : : "m" (dptr[z][d]));
+			asm volatile("prefetchnta %0" : : "m" (dptr[z][d+32]));
+			asm volatile("prefetchnta %0" : : "m" (dptr[z][d+64]));
+			asm volatile("prefetchnta %0" : : "m" (dptr[z][d+96]));
+			asm volatile("vpcmpgtb %ymm4,%ymm1,%ymm5");
+			asm volatile("vpcmpgtb %ymm6,%ymm1,%ymm7");
+			asm volatile("vpcmpgtb %ymm12,%ymm1,%ymm13");
+			asm volatile("vpcmpgtb %ymm14,%ymm1,%ymm15");
+			asm volatile("vpaddb %ymm4,%ymm4,%ymm4");
+			asm volatile("vpaddb %ymm6,%ymm6,%ymm6");
+			asm volatile("vpaddb %ymm12,%ymm12,%ymm12");
+			asm volatile("vpaddb %ymm14,%ymm14,%ymm14");
+			asm volatile("vpand %ymm0,%ymm5,%ymm5");
+			asm volatile("vpand %ymm0,%ymm7,%ymm7");
+			asm volatile("vpand %ymm0,%ymm13,%ymm13");
+			asm volatile("vpand %ymm0,%ymm15,%ymm15");
+			asm volatile("vpxor %ymm5,%ymm4,%ymm4");
+			asm volatile("vpxor %ymm7,%ymm6,%ymm6");
+			asm volatile("vpxor %ymm13,%ymm12,%ymm12");
+			asm volatile("vpxor %ymm15,%ymm14,%ymm14");
+			asm volatile("vmovdqa %0,%%ymm5" : : "m" (dptr[z][d]));
+			asm volatile("vmovdqa %0,%%ymm7" : : "m" (dptr[z][d+32]));
+			asm volatile("vmovdqa %0,%%ymm13" : : "m" (dptr[z][d+64]));
+			asm volatile("vmovdqa %0,%%ymm15" : : "m" (dptr[z][d+96]));
+			asm volatile("vpxor %ymm5,%ymm2,%ymm2");
+			asm volatile("vpxor %ymm7,%ymm3,%ymm3");
+			asm volatile("vpxor %ymm13,%ymm10,%ymm10");
+			asm volatile("vpxor %ymm15,%ymm11,%ymm11");
+			asm volatile("vpxor %ymm5,%ymm4,%ymm4");
+			asm volatile("vpxor %ymm7,%ymm6,%ymm6");
+			asm volatile("vpxor %ymm13,%ymm12,%ymm12");
+			asm volatile("vpxor %ymm15,%ymm14,%ymm14");
+		}
+		asm volatile("vmovntdq %%ymm2,%0" : "=m" (p[d]));
+		asm volatile("vpxor %ymm2,%ymm2,%ymm2");
+		asm volatile("vmovntdq %%ymm3,%0" : "=m" (p[d+32]));
+		asm volatile("vpxor %ymm3,%ymm3,%ymm3");
+		asm volatile("vmovntdq %%ymm10,%0" : "=m" (p[d+64]));
+		asm volatile("vpxor %ymm10,%ymm10,%ymm10");
+		asm volatile("vmovntdq %%ymm11,%0" : "=m" (p[d+96]));
+		asm volatile("vpxor %ymm11,%ymm11,%ymm11");
+		asm volatile("vmovntdq %%ymm4,%0" : "=m" (q[d]));
+		asm volatile("vpxor %ymm4,%ymm4,%ymm4");
+		asm volatile("vmovntdq %%ymm6,%0" : "=m" (q[d+32]));
+		asm volatile("vpxor %ymm6,%ymm6,%ymm6");
+		asm volatile("vmovntdq %%ymm12,%0" : "=m" (q[d+64]));
+		asm volatile("vpxor %ymm12,%ymm12,%ymm12");
+		asm volatile("vmovntdq %%ymm14,%0" : "=m" (q[d+96]));
+		asm volatile("vpxor %ymm14,%ymm14,%ymm14");
+	}
+
+	asm volatile("sfence" : : : "memory");
+	kernel_fpu_end();
+}
+
+const struct raid6_calls raid6_avx2x4 = {
+	raid6_avx24_gen_syndrome,
+	raid6_have_avx2,
+	"avx2x4",
+	1			/* Has cache hints */
+};
+#endif
+
+#endif /* CONFIG_AS_AVX2 */
diff --git a/lib/raid6/test/Makefile b/lib/raid6/test/Makefile
index d919c98..754cbac 100644
--- a/lib/raid6/test/Makefile
+++ b/lib/raid6/test/Makefile
@@ -11,6 +11,16 @@ AWK	 = awk -f
 AR	 = ar
 RANLIB	 = ranlib
 
+ARCH := $(shell uname -m 2>/dev/null | sed -e /s/i.86/i386/)
+ifeq ($(ARCH),i386)
+        CFLAGS += -DCONFIG_X86_32
+endif
+ifeq ($(ARCH),x86_64)
+        CFLAGS += -DCONFIG_X86_64
+endif
+CFLAGS += $(shell echo "vpbroadcastb %xmm0, %ymm1"| gcc -c -x assembler - &&\
+	     rm ./-.o && echo -DCONFIG_AS_AVX2=1)
+
 .c.o:
 	$(CC) $(CFLAGS) -c -o $@ $<
 
@@ -22,7 +32,7 @@ RANLIB	 = ranlib
 
 all:	raid6.a raid6test
 
-raid6.a: int1.o int2.o int4.o int8.o int16.o int32.o mmx.o sse1.o sse2.o \
+raid6.a: int1.o int2.o int4.o int8.o int16.o int32.o mmx.o sse1.o sse2.o avx2.o \
 	 altivec1.o altivec2.o altivec4.o altivec8.o recov.o recov_ssse3.o recov_avx2.o algos.o \
 	 tables.o
 	 rm -f $@