Adapt jxiong:vectorized_fletcher (openzfs#4330) to

openzfs#4381
ironMann · Mar 16, 2016 · 904baa9 · 904baa9
1 parent b378ac3
commit 904baa9
Show file tree

Hide file tree

Showing 4 changed files with 25 additions and 64 deletions.
diff --git a/include/sys/Makefile.am b/include/sys/Makefile.am
@@ -104,8 +104,7 @@ KERNEL_H = \
 	$(top_srcdir)/include/sys/zfs_ioctl.h \
 	$(top_srcdir)/include/sys/zfs_onexit.h \
 	${top_srcdir}/include/sys/zpl.h \
-	$(top_srcdir)/include/sys/zvol.h \
-	$(top_srcdir)/include/sys/platform_cpu_compat.h
+	$(top_srcdir)/include/sys/zvol.h
 
 USER_H =
 

diff --git a/include/sys/platform_cpu_compat.h b/include/sys/platform_cpu_compat.h
diff --git a/module/zcommon/zfs_fletcher.c b/module/zcommon/zfs_fletcher.c
@@ -128,8 +128,6 @@
 #include <sys/types.h>
 #include <sys/sysmacros.h>
 #include <sys/byteorder.h>
-#include <sys/zio.h>
-#include <sys/spa.h>
 #include <zfs_fletcher.h>
 
 void
@@ -256,10 +254,10 @@ fletcher_4_incremental_byteswap(const void *buf, uint64_t size,
 	fletcher_4_generic_byteswap(buf, size, zcp);
 }
 
+
 #if defined(_KERNEL) && defined(HAVE_SPL)
-#include <sys/platform_cpu_compat.h>
 
-#ifdef HAVE_KERNEL_CPU_AVX2
+#if defined(HAVE_AVX) && defined(HAVE_AVX2)
 #include "zfs_fletcher_intel.c"
 #endif
 
@@ -270,15 +268,21 @@ static const struct fletcher_4_calls *fletcher_4_algos[] = {
 #endif
 };
 
+#define	BENCH_SIZE 4096
+#define	ZFS_ARRAY_SIZE(a) (sizeof (a) / sizeof (*a))
+#define	kernel_cpu_relax() do {} while (0)
+
+/* cant use allocation methods from zfs module! */
+static char databuf[BENCH_SIZE];
+
 void
 fletcher_4_init(void)
 {
 	unsigned long bestperf = 0;
-	const void *databuf = current_text_addr();
 	const unsigned int bits = 4;
 	int i;
 
-	for (i = 0; i < ARRAY_SIZE(fletcher_4_algos); i++) {
+	for (i = 0; i < ZFS_ARRAY_SIZE(fletcher_4_algos); i++) {
 		const struct fletcher_4_calls *algo = fletcher_4_algos[i];
 		unsigned long perf = 0;
 		clock_t j0, j1;
@@ -288,12 +292,13 @@ fletcher_4_init(void)
 			continue;
 
 		kpreempt_disable();
-		j0 = lbolt;
-		while ((j1 = lbolt) == j0)
+		j0 = ddi_get_lbolt();
+		while ((j1 = ddi_get_lbolt()) == j0) {
 			kernel_cpu_relax();
+		}
 
 		algo->init(&zc);
-		while (ddi_time_before(lbolt, j1 + (1 << bits))) {
+		while (ddi_time_before(ddi_get_lbolt(), j1 + (1 << bits))) {
 			algo->compute(databuf, PAGE_SIZE, &zc);
 			perf++;
 		}

diff --git a/module/zcommon/zfs_fletcher_intel.c b/module/zcommon/zfs_fletcher_intel.c
@@ -40,6 +40,8 @@
  * SOFTWARE.
  */
 
+#if defined(HAVE_AVX) && defined(HAVE_AVX2)
+
 #ifdef UNITTEST
 
 #if defined(_KERNEL)
@@ -54,28 +56,27 @@
 #include <stdbool.h>
 #include <string.h>
 
-#define	kernel_fpu_save()	do {} while (0)
-#define	kernel_fpu_restore()	do {} while (0)
-#define	cpu_has_avx2 __builtin_cpu_supports("avx2")
-
 typedef unsigned long long rlim64_t;
 
 #include <sys/types.h>
 #include <sys/sysmacros.h>
 #include <sys/byteorder.h>
 #include <sys/zio.h>
 #include <sys/spa.h>
+
 #include <zfs_fletcher.h>
 
 #include "zfs_fletcher.c"
 #endif /* #ifdef UNITTEST */
 
+#include <linux/simd_x86.h>
+
 static void
 fletcher_4_avx2_init(zio_cksum_t *zcp)
 {
 	ZIO_SET_CHECKSUM(zcp, 0, 0, 0, 0);
 
-	kernel_fpu_save();
+	kfpu_begin();
 
 	/* clear avx2 registers */
 	asm volatile("vpxor %ymm0, %ymm0, %ymm0");
@@ -98,7 +99,7 @@ fletcher_4_avx2_fini(zio_cksum_t *zcp)
 	asm volatile("vmovdqa %%ymm2, %0":"=m" (c));
 	asm volatile("vmovdqa %%ymm3, %0":"=m" (d));
 
-	kernel_fpu_restore();
+	kfpu_end();
 
 	A = a[0] + a[1] + a[2] + a[3];
 	B = 0 - a[1] - 2*a[2] - 3*a[3]
@@ -155,7 +156,7 @@ fletcher_4_avx2_byteswap(const void *buf, uint64_t size, zio_cksum_t *unused)
 
 static boolean_t fletcher_4_avx2_valid(void)
 {
-	return (cpu_has_avx2);
+	return (zfs_avx_available() && zfs_avx2_available());
 }
 
 static const struct fletcher_4_calls fletcher_4_avx2_calls = {
@@ -249,3 +250,5 @@ main(int argc, char **argv)
 	return (0);
 }
 #endif /* #ifdef UNITTEST */
+
+#endif /* defined(HAVE_AVX) && defined(HAVE_AVX2) */