From 7d1353fa01da000945f747bd8d26673e2745487c Mon Sep 17 00:00:00 2001
From: David Sterba <dsterba@suse.com>
Date: Thu, 9 Feb 2023 02:54:35 +0100
Subject: [PATCH] btrfs-progs: hash-speedtest: add accelerated BLAKE2
 implementations

Benchmark all accelerated implementations if the CPU supports them. Set
the level before each test, expecting that the implementation switches
the implementation dynamically.

Signed-off-by: David Sterba <dsterba@suse.com>
---
 crypto/hash-speedtest.c | 19 ++++++++++++++++++-
 1 file changed, 18 insertions(+), 1 deletion(-)

diff --git a/crypto/hash-speedtest.c b/crypto/hash-speedtest.c
index 21e7697e..83217b17 100644
--- a/crypto/hash-speedtest.c
+++ b/crypto/hash-speedtest.c
@@ -29,6 +29,7 @@
 #include "crypto/sha.h"
 #include "crypto/blake2.h"
 #include "common/messages.h"
+#include "common/cpu-utils.h"
 
 #ifdef __x86_64__
 static const int cycles_supported = 1;
@@ -181,16 +182,26 @@ int main(int argc, char **argv) {
 		int digest_size;
 		u64 cycles;
 		u64 time;
+		unsigned long cpu_flag;
 	} contestants[] = {
 		{ .name = "NULL-NOP", .digest = hash_null_nop, .digest_size = 32 },
 		{ .name = "NULL-MEMCPY", .digest = hash_null_memcpy, .digest_size = 32 },
 		{ .name = "CRC32C", .digest = hash_crc32c, .digest_size = 4 },
 		{ .name = "XXHASH", .digest = hash_xxhash, .digest_size = 8 },
 		{ .name = "SHA256", .digest = hash_sha256, .digest_size = 32 },
-		{ .name = "BLAKE2", .digest = hash_blake2b, .digest_size = 32 },
+		{ .name = "BLAKE2-ref", .digest = hash_blake2b, .digest_size = 32 },
+		{ .name = "BLAKE2-SSE2", .digest = hash_blake2b, .digest_size = 32,
+		  .cpu_flag = CPU_FLAG_SSE2 },
+		{ .name = "BLAKE2-SSE41", .digest = hash_blake2b, .digest_size = 32,
+		  .cpu_flag = CPU_FLAG_SSE41 },
+		{ .name = "BLAKE2-AVX2", .digest = hash_blake2b, .digest_size = 32,
+		  .cpu_flag = CPU_FLAG_AVX2 },
 	};
 	int units = UNITS_CYCLES;
 
+	cpu_detect_flags();
+	cpu_print_flags();
+
 	optind = 0;
 	while (1) {
 		static const struct option long_options[] = {
@@ -250,9 +261,14 @@ int main(int argc, char **argv) {
 		u64 tstart, tend;
 		u64 total = 0;
 
+		if (c->cpu_flag != 0 && !cpu_has_feature(c->cpu_flag)) {
+			printf("%12s: no CPU support\n", c->name);
+			continue;
+		}
 		printf("%12s: ", c->name);
 		fflush(stdout);
 
+		cpu_set_level(c->cpu_flag);
 		tstart = get_time();
 		start = get_cycles(units);
 		for (iter = 0; iter < iterations; iter++) {
@@ -264,6 +280,7 @@ int main(int argc, char **argv) {
 		tend = get_time();
 		c->cycles = end - start;
 		c->time = tend - tstart;
+		cpu_reset_level();
 
 		if (units == UNITS_CYCLES || units == UNITS_PERF)
 			total = c->cycles;