summaryrefslogtreecommitdiff
path: root/lib/raid6/mmx.c
diff options
context:
space:
mode:
authorDavid Woodhouse <David.Woodhouse@intel.com>2010-08-11 00:19:05 +0100
committerDavid Woodhouse <David.Woodhouse@intel.com>2010-08-11 00:19:05 +0100
commitcc4589ebfae6f8dbb5cf880a0a67eedab3416492 (patch)
tree404335e6dec9d79351f537de58979a8b964e2b0c /lib/raid6/mmx.c
parent3d30701b58970425e1d45994d6cb82f828924fdd (diff)
Rename raid6 files now they're in a 'raid6' directory.
Linus asks 'why "raid6" twice?'. No reason. Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Diffstat (limited to 'lib/raid6/mmx.c')
-rw-r--r--lib/raid6/mmx.c142
1 files changed, 142 insertions, 0 deletions
diff --git a/lib/raid6/mmx.c b/lib/raid6/mmx.c
new file mode 100644
index 000000000000..e7f6c13132bf
--- /dev/null
+++ b/lib/raid6/mmx.c
@@ -0,0 +1,142 @@
+/* -*- linux-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.
+ *
+ * ----------------------------------------------------------------------- */
+
+/*
+ * raid6mmx.c
+ *
+ * MMX implementation of RAID-6 syndrome functions
+ */
+
+#if defined(__i386__) && !defined(__arch_um__)
+
+#include <linux/raid/pq.h>
+#include "raid6x86.h"
+
+/* Shared with raid6sse1.c */
+const struct raid6_mmx_constants {
+ u64 x1d;
+} raid6_mmx_constants = {
+ 0x1d1d1d1d1d1d1d1dULL,
+};
+
+static int raid6_have_mmx(void)
+{
+ /* Not really "boot_cpu" but "all_cpus" */
+ return boot_cpu_has(X86_FEATURE_MMX);
+}
+
+/*
+ * Plain MMX implementation
+ */
+static void raid6_mmx1_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("movq %0,%%mm0" : : "m" (raid6_mmx_constants.x1d));
+ asm volatile("pxor %mm5,%mm5"); /* Zero temp */
+
+ for ( d = 0 ; d < bytes ; d += 8 ) {
+ asm volatile("movq %0,%%mm2" : : "m" (dptr[z0][d])); /* P[0] */
+ asm volatile("movq %mm2,%mm4"); /* Q[0] */
+ for ( z = z0-1 ; z >= 0 ; z-- ) {
+ asm volatile("movq %0,%%mm6" : : "m" (dptr[z][d]));
+ asm volatile("pcmpgtb %mm4,%mm5");
+ asm volatile("paddb %mm4,%mm4");
+ asm volatile("pand %mm0,%mm5");
+ asm volatile("pxor %mm5,%mm4");
+ asm volatile("pxor %mm5,%mm5");
+ asm volatile("pxor %mm6,%mm2");
+ asm volatile("pxor %mm6,%mm4");
+ }
+ asm volatile("movq %%mm2,%0" : "=m" (p[d]));
+ asm volatile("pxor %mm2,%mm2");
+ asm volatile("movq %%mm4,%0" : "=m" (q[d]));
+ asm volatile("pxor %mm4,%mm4");
+ }
+
+ kernel_fpu_end();
+}
+
+const struct raid6_calls raid6_mmxx1 = {
+ raid6_mmx1_gen_syndrome,
+ raid6_have_mmx,
+ "mmxx1",
+ 0
+};
+
+/*
+ * Unrolled-by-2 MMX implementation
+ */
+static void raid6_mmx2_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("movq %0,%%mm0" : : "m" (raid6_mmx_constants.x1d));
+ asm volatile("pxor %mm5,%mm5"); /* Zero temp */
+ asm volatile("pxor %mm7,%mm7"); /* Zero temp */
+
+ for ( d = 0 ; d < bytes ; d += 16 ) {
+ asm volatile("movq %0,%%mm2" : : "m" (dptr[z0][d])); /* P[0] */
+ asm volatile("movq %0,%%mm3" : : "m" (dptr[z0][d+8]));
+ asm volatile("movq %mm2,%mm4"); /* Q[0] */
+ asm volatile("movq %mm3,%mm6"); /* Q[1] */
+ for ( z = z0-1 ; z >= 0 ; z-- ) {
+ asm volatile("pcmpgtb %mm4,%mm5");
+ asm volatile("pcmpgtb %mm6,%mm7");
+ asm volatile("paddb %mm4,%mm4");
+ asm volatile("paddb %mm6,%mm6");
+ asm volatile("pand %mm0,%mm5");
+ asm volatile("pand %mm0,%mm7");
+ asm volatile("pxor %mm5,%mm4");
+ asm volatile("pxor %mm7,%mm6");
+ asm volatile("movq %0,%%mm5" : : "m" (dptr[z][d]));
+ asm volatile("movq %0,%%mm7" : : "m" (dptr[z][d+8]));
+ asm volatile("pxor %mm5,%mm2");
+ asm volatile("pxor %mm7,%mm3");
+ asm volatile("pxor %mm5,%mm4");
+ asm volatile("pxor %mm7,%mm6");
+ asm volatile("pxor %mm5,%mm5");
+ asm volatile("pxor %mm7,%mm7");
+ }
+ asm volatile("movq %%mm2,%0" : "=m" (p[d]));
+ asm volatile("movq %%mm3,%0" : "=m" (p[d+8]));
+ asm volatile("movq %%mm4,%0" : "=m" (q[d]));
+ asm volatile("movq %%mm6,%0" : "=m" (q[d+8]));
+ }
+
+ kernel_fpu_end();
+}
+
+const struct raid6_calls raid6_mmxx2 = {
+ raid6_mmx2_gen_syndrome,
+ raid6_have_mmx,
+ "mmxx2",
+ 0
+};
+
+#endif