Import blake2sjd/blake2s-comparison

6 files changed, 847 insertions, 501 deletions

diff --git a/Makefile b/Makefile
index 1c9e6ec..69e8157 100644
--- a/Makefile
+++ b/Makefile
@@ -1,5 +1,5 @@
 ifneq ($(KERNELRELEASE),)
-kbench9000-y := main.o curve25519-u128.o
+kbench9000-y := main.o blake2s-compress.o
 obj-m := kbench9000.o
 ccflags-y += -O3
 ccflags-y += -D'pr_fmt(fmt)=KBUILD_MODNAME ": " fmt'
diff --git a/blake2s-compress.S b/blake2s-compress.S
new file mode 100644
index 0000000..5b67a54
--- /dev/null
+++ b/blake2s-compress.S
@@ -0,0 +1,762 @@
+/* SPDX-License-Identifier: GPL-2.0 OR MIT */
+/*
+ * Copyright (C) 2015-2018 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved.
+ * Copyright (C) 2017 Samuel Neves <sneves@dei.uc.pt>. All Rights Reserved.
+ */
+
+#include <linux/linkage.h>
+
+.section .rodata.cst32.BLAKE2S_IV, "aM", @progbits, 32
+.align 32
+IV:	.octa 0xA54FF53A3C6EF372BB67AE856A09E667
+	.octa 0x5BE0CD191F83D9AB9B05688C510E527F
+.section .rodata.cst16.ROT16, "aM", @progbits, 16
+.align 16
+ROT16:	.octa 0x0D0C0F0E09080B0A0504070601000302
+.section .rodata.cst16.ROR328, "aM", @progbits, 16
+.align 16
+ROR328:	.octa 0x0C0F0E0D080B0A090407060500030201
+#ifdef CONFIG_AS_AVX512
+.section .rodata.cst64.BLAKE2S_SIGMA, "aM", @progbits, 640
+.align 64
+SIGMA:
+.long 0, 2, 4, 6, 1, 3, 5, 7, 8, 10, 12, 14, 9, 11, 13, 15
+.long 11, 2, 12, 14, 9, 8, 15, 3, 4, 0, 13, 6, 10, 1, 7, 5
+.long 10, 12, 11, 6, 5, 9, 13, 3, 4, 15, 14, 2, 0, 7, 8, 1
+.long 10, 9, 7, 0, 11, 14, 1, 12, 6, 2, 15, 3, 13, 8, 5, 4
+.long 4, 9, 8, 13, 14, 0, 10, 11, 7, 3, 12, 1, 5, 6, 15, 2
+.long 2, 10, 4, 14, 13, 3, 9, 11, 6, 5, 7, 12, 15, 1, 8, 0
+.long 4, 11, 14, 8, 13, 10, 12, 5, 2, 1, 15, 3, 9, 7, 0, 6
+.long 6, 12, 0, 13, 15, 2, 1, 10, 4, 5, 11, 14, 8, 3, 9, 7
+.long 14, 5, 4, 12, 9, 7, 3, 10, 2, 0, 6, 15, 11, 1, 13, 8
+.long 11, 7, 13, 10, 12, 14, 0, 15, 4, 5, 6, 9, 2, 1, 8, 3
+#endif /* CONFIG_AS_AVX512 */
+
+.text
+#ifdef CONFIG_AS_AVX
+ENTRY(blake2s_compress_avx)
+	movl		%ecx, %ecx
+	testq		%rdx, %rdx
+	je		.Lendofloop
+	.align 32
+.Lbeginofloop:
+	addq		%rcx, 32(%rdi)
+	vmovdqu		IV+16(%rip), %xmm1
+	vmovdqu		(%rsi), %xmm4
+	vpxor		32(%rdi), %xmm1, %xmm1
+	vmovdqu		16(%rsi), %xmm3
+	vshufps		$136, %xmm3, %xmm4, %xmm6
+	vmovdqa		ROT16(%rip), %xmm7
+	vpaddd		(%rdi), %xmm6, %xmm6
+	vpaddd		16(%rdi), %xmm6, %xmm6
+	vpxor		%xmm6, %xmm1, %xmm1
+	vmovdqu		IV(%rip), %xmm8
+	vpshufb		%xmm7, %xmm1, %xmm1
+	vmovdqu		48(%rsi), %xmm5
+	vpaddd		%xmm1, %xmm8, %xmm8
+	vpxor		16(%rdi), %xmm8, %xmm9
+	vmovdqu		32(%rsi), %xmm2
+	vpblendw	$12, %xmm3, %xmm5, %xmm13
+	vshufps		$221, %xmm5, %xmm2, %xmm12
+	vpunpckhqdq	%xmm2, %xmm4, %xmm14
+	vpslld		$20, %xmm9, %xmm0
+	vpsrld		$12, %xmm9, %xmm9
+	vpxor		%xmm0, %xmm9, %xmm0
+	vshufps		$221, %xmm3, %xmm4, %xmm9
+	vpaddd		%xmm9, %xmm6, %xmm9
+	vpaddd		%xmm0, %xmm9, %xmm9
+	vpxor		%xmm9, %xmm1, %xmm1
+	vmovdqa		ROR328(%rip), %xmm6
+	vpshufb		%xmm6, %xmm1, %xmm1
+	vpaddd		%xmm1, %xmm8, %xmm8
+	vpxor		%xmm8, %xmm0, %xmm0
+	vpshufd		$147, %xmm1, %xmm1
+	vpshufd		$78, %xmm8, %xmm8
+	vpslld		$25, %xmm0, %xmm10
+	vpsrld		$7, %xmm0, %xmm0
+	vpxor		%xmm10, %xmm0, %xmm0
+	vshufps		$136, %xmm5, %xmm2, %xmm10
+	vpshufd		$57, %xmm0, %xmm0
+	vpaddd		%xmm10, %xmm9, %xmm9
+	vpaddd		%xmm0, %xmm9, %xmm9
+	vpxor		%xmm9, %xmm1, %xmm1
+	vpaddd		%xmm12, %xmm9, %xmm9
+	vpblendw	$12, %xmm2, %xmm3, %xmm12
+	vpshufb		%xmm7, %xmm1, %xmm1
+	vpaddd		%xmm1, %xmm8, %xmm8
+	vpxor		%xmm8, %xmm0, %xmm10
+	vpslld		$20, %xmm10, %xmm0
+	vpsrld		$12, %xmm10, %xmm10
+	vpxor		%xmm0, %xmm10, %xmm0
+	vpaddd		%xmm0, %xmm9, %xmm9
+	vpxor		%xmm9, %xmm1, %xmm1
+	vpshufb		%xmm6, %xmm1, %xmm1
+	vpaddd		%xmm1, %xmm8, %xmm8
+	vpxor		%xmm8, %xmm0, %xmm0
+	vpshufd		$57, %xmm1, %xmm1
+	vpshufd		$78, %xmm8, %xmm8
+	vpslld		$25, %xmm0, %xmm10
+	vpsrld		$7, %xmm0, %xmm0
+	vpxor		%xmm10, %xmm0, %xmm0
+	vpslldq		$4, %xmm5, %xmm10
+	vpblendw	$240, %xmm10, %xmm12, %xmm12
+	vpshufd		$147, %xmm0, %xmm0
+	vpshufd		$147, %xmm12, %xmm12
+	vpaddd		%xmm9, %xmm12, %xmm12
+	vpaddd		%xmm0, %xmm12, %xmm12
+	vpxor		%xmm12, %xmm1, %xmm1
+	vpshufb		%xmm7, %xmm1, %xmm1
+	vpaddd		%xmm1, %xmm8, %xmm8
+	vpxor		%xmm8, %xmm0, %xmm11
+	vpslld		$20, %xmm11, %xmm9
+	vpsrld		$12, %xmm11, %xmm11
+	vpxor		%xmm9, %xmm11, %xmm0
+	vpshufd		$8, %xmm2, %xmm9
+	vpblendw	$192, %xmm5, %xmm3, %xmm11
+	vpblendw	$240, %xmm11, %xmm9, %xmm9
+	vpshufd		$177, %xmm9, %xmm9
+	vpaddd		%xmm12, %xmm9, %xmm9
+	vpaddd		%xmm0, %xmm9, %xmm11
+	vpxor		%xmm11, %xmm1, %xmm1
+	vpshufb		%xmm6, %xmm1, %xmm1
+	vpaddd		%xmm1, %xmm8, %xmm8
+	vpxor		%xmm8, %xmm0, %xmm9
+	vpshufd		$147, %xmm1, %xmm1
+	vpshufd		$78, %xmm8, %xmm8
+	vpslld		$25, %xmm9, %xmm0
+	vpsrld		$7, %xmm9, %xmm9
+	vpxor		%xmm0, %xmm9, %xmm0
+	vpslldq		$4, %xmm3, %xmm9
+	vpblendw	$48, %xmm9, %xmm2, %xmm9
+	vpblendw	$240, %xmm9, %xmm4, %xmm9
+	vpshufd		$57, %xmm0, %xmm0
+	vpshufd		$177, %xmm9, %xmm9
+	vpaddd		%xmm11, %xmm9, %xmm9
+	vpaddd		%xmm0, %xmm9, %xmm9
+	vpxor		%xmm9, %xmm1, %xmm1
+	vpshufb		%xmm7, %xmm1, %xmm1
+	vpaddd		%xmm1, %xmm8, %xmm11
+	vpxor		%xmm11, %xmm0, %xmm0
+	vpslld		$20, %xmm0, %xmm8
+	vpsrld		$12, %xmm0, %xmm0
+	vpxor		%xmm8, %xmm0, %xmm0
+	vpunpckhdq	%xmm3, %xmm4, %xmm8
+	vpblendw	$12, %xmm10, %xmm8, %xmm12
+	vpshufd		$177, %xmm12, %xmm12
+	vpaddd		%xmm9, %xmm12, %xmm9
+	vpaddd		%xmm0, %xmm9, %xmm9
+	vpxor		%xmm9, %xmm1, %xmm1
+	vpshufb		%xmm6, %xmm1, %xmm1
+	vpaddd		%xmm1, %xmm11, %xmm11
+	vpxor		%xmm11, %xmm0, %xmm0
+	vpshufd		$57, %xmm1, %xmm1
+	vpshufd		$78, %xmm11, %xmm11
+	vpslld		$25, %xmm0, %xmm12
+	vpsrld		$7, %xmm0, %xmm0
+	vpxor		%xmm12, %xmm0, %xmm0
+	vpunpckhdq	%xmm5, %xmm2, %xmm12
+	vpshufd		$147, %xmm0, %xmm0
+	vpblendw	$15, %xmm13, %xmm12, %xmm12
+	vpslldq		$8, %xmm5, %xmm13
+	vpshufd		$210, %xmm12, %xmm12
+	vpaddd		%xmm9, %xmm12, %xmm9
+	vpaddd		%xmm0, %xmm9, %xmm9
+	vpxor		%xmm9, %xmm1, %xmm1
+	vpshufb		%xmm7, %xmm1, %xmm1
+	vpaddd		%xmm1, %xmm11, %xmm11
+	vpxor		%xmm11, %xmm0, %xmm0
+	vpslld		$20, %xmm0, %xmm12
+	vpsrld		$12, %xmm0, %xmm0
+	vpxor		%xmm12, %xmm0, %xmm0
+	vpunpckldq	%xmm4, %xmm2, %xmm12
+	vpblendw	$240, %xmm4, %xmm12, %xmm12
+	vpblendw	$192, %xmm13, %xmm12, %xmm12
+	vpsrldq		$12, %xmm3, %xmm13
+	vpaddd		%xmm12, %xmm9, %xmm9
+	vpaddd		%xmm0, %xmm9, %xmm9
+	vpxor		%xmm9, %xmm1, %xmm1
+	vpshufb		%xmm6, %xmm1, %xmm1
+	vpaddd		%xmm1, %xmm11, %xmm11
+	vpxor		%xmm11, %xmm0, %xmm0
+	vpshufd		$147, %xmm1, %xmm1
+	vpshufd		$78, %xmm11, %xmm11
+	vpslld		$25, %xmm0, %xmm12
+	vpsrld		$7, %xmm0, %xmm0
+	vpxor		%xmm12, %xmm0, %xmm0
+	vpblendw	$60, %xmm2, %xmm4, %xmm12
+	vpblendw	$3, %xmm13, %xmm12, %xmm12
+	vpshufd		$57, %xmm0, %xmm0
+	vpshufd		$78, %xmm12, %xmm12
+	vpaddd		%xmm9, %xmm12, %xmm9
+	vpaddd		%xmm0, %xmm9, %xmm9
+	vpxor		%xmm9, %xmm1, %xmm1
+	vpshufb		%xmm7, %xmm1, %xmm1
+	vpaddd		%xmm1, %xmm11, %xmm11
+	vpxor		%xmm11, %xmm0, %xmm12
+	vpslld		$20, %xmm12, %xmm13
+	vpsrld		$12, %xmm12, %xmm0
+	vpblendw	$51, %xmm3, %xmm4, %xmm12
+	vpxor		%xmm13, %xmm0, %xmm0
+	vpblendw	$192, %xmm10, %xmm12, %xmm10
+	vpslldq		$8, %xmm2, %xmm12
+	vpshufd		$27, %xmm10, %xmm10
+	vpaddd		%xmm9, %xmm10, %xmm9
+	vpaddd		%xmm0, %xmm9, %xmm9
+	vpxor		%xmm9, %xmm1, %xmm1
+	vpshufb		%xmm6, %xmm1, %xmm1
+	vpaddd		%xmm1, %xmm11, %xmm11
+	vpxor		%xmm11, %xmm0, %xmm0
+	vpshufd		$57, %xmm1, %xmm1
+	vpshufd		$78, %xmm11, %xmm11
+	vpslld		$25, %xmm0, %xmm10
+	vpsrld		$7, %xmm0, %xmm0
+	vpxor		%xmm10, %xmm0, %xmm0
+	vpunpckhdq	%xmm2, %xmm8, %xmm10
+	vpshufd		$147, %xmm0, %xmm0
+	vpblendw	$12, %xmm5, %xmm10, %xmm10
+	vpshufd		$210, %xmm10, %xmm10
+	vpaddd		%xmm9, %xmm10, %xmm9
+	vpaddd		%xmm0, %xmm9, %xmm9
+	vpxor		%xmm9, %xmm1, %xmm1
+	vpshufb		%xmm7, %xmm1, %xmm1
+	vpaddd		%xmm1, %xmm11, %xmm11
+	vpxor		%xmm11, %xmm0, %xmm10
+	vpslld		$20, %xmm10, %xmm0
+	vpsrld		$12, %xmm10, %xmm10
+	vpxor		%xmm0, %xmm10, %xmm0
+	vpblendw	$12, %xmm4, %xmm5, %xmm10
+	vpblendw	$192, %xmm12, %xmm10, %xmm10
+	vpunpckldq	%xmm2, %xmm4, %xmm12
+	vpshufd		$135, %xmm10, %xmm10
+	vpaddd		%xmm9, %xmm10, %xmm9
+	vpaddd		%xmm0, %xmm9, %xmm9
+	vpxor		%xmm9, %xmm1, %xmm1
+	vpshufb		%xmm6, %xmm1, %xmm1
+	vpaddd		%xmm1, %xmm11, %xmm13
+	vpxor		%xmm13, %xmm0, %xmm0
+	vpshufd		$147, %xmm1, %xmm1
+	vpshufd		$78, %xmm13, %xmm13
+	vpslld		$25, %xmm0, %xmm10
+	vpsrld		$7, %xmm0, %xmm0
+	vpxor		%xmm10, %xmm0, %xmm0
+	vpblendw	$15, %xmm3, %xmm4, %xmm10
+	vpblendw	$192, %xmm5, %xmm10, %xmm10
+	vpshufd		$57, %xmm0, %xmm0
+	vpshufd		$198, %xmm10, %xmm10
+	vpaddd		%xmm9, %xmm10, %xmm10
+	vpaddd		%xmm0, %xmm10, %xmm10
+	vpxor		%xmm10, %xmm1, %xmm1
+	vpshufb		%xmm7, %xmm1, %xmm1
+	vpaddd		%xmm1, %xmm13, %xmm13
+	vpxor		%xmm13, %xmm0, %xmm9
+	vpslld		$20, %xmm9, %xmm0
+	vpsrld		$12, %xmm9, %xmm9
+	vpxor		%xmm0, %xmm9, %xmm0
+	vpunpckhdq	%xmm2, %xmm3, %xmm9
+	vpunpcklqdq	%xmm12, %xmm9, %xmm15
+	vpunpcklqdq	%xmm12, %xmm8, %xmm12
+	vpblendw	$15, %xmm5, %xmm8, %xmm8
+	vpaddd		%xmm15, %xmm10, %xmm15
+	vpaddd		%xmm0, %xmm15, %xmm15
+	vpxor		%xmm15, %xmm1, %xmm1
+	vpshufd		$141, %xmm8, %xmm8
+	vpshufb		%xmm6, %xmm1, %xmm1
+	vpaddd		%xmm1, %xmm13, %xmm13
+	vpxor		%xmm13, %xmm0, %xmm0
+	vpshufd		$57, %xmm1, %xmm1
+	vpshufd		$78, %xmm13, %xmm13
+	vpslld		$25, %xmm0, %xmm10
+	vpsrld		$7, %xmm0, %xmm0
+	vpxor		%xmm10, %xmm0, %xmm0
+	vpunpcklqdq	%xmm2, %xmm3, %xmm10
+	vpshufd		$147, %xmm0, %xmm0
+	vpblendw	$51, %xmm14, %xmm10, %xmm14
+	vpshufd		$135, %xmm14, %xmm14
+	vpaddd		%xmm15, %xmm14, %xmm14
+	vpaddd		%xmm0, %xmm14, %xmm14
+	vpxor		%xmm14, %xmm1, %xmm1
+	vpunpcklqdq	%xmm3, %xmm4, %xmm15
+	vpshufb		%xmm7, %xmm1, %xmm1
+	vpaddd		%xmm1, %xmm13, %xmm13
+	vpxor		%xmm13, %xmm0, %xmm0
+	vpslld		$20, %xmm0, %xmm11
+	vpsrld		$12, %xmm0, %xmm0
+	vpxor		%xmm11, %xmm0, %xmm0
+	vpunpckhqdq	%xmm5, %xmm3, %xmm11
+	vpblendw	$51, %xmm15, %xmm11, %xmm11
+	vpunpckhqdq	%xmm3, %xmm5, %xmm15
+	vpaddd		%xmm11, %xmm14, %xmm11
+	vpaddd		%xmm0, %xmm11, %xmm11
+	vpxor		%xmm11, %xmm1, %xmm1
+	vpshufb		%xmm6, %xmm1, %xmm1
+	vpaddd		%xmm1, %xmm13, %xmm13
+	vpxor		%xmm13, %xmm0, %xmm0
+	vpshufd		$147, %xmm1, %xmm1
+	vpshufd		$78, %xmm13, %xmm13
+	vpslld		$25, %xmm0, %xmm14
+	vpsrld		$7, %xmm0, %xmm0
+	vpxor		%xmm14, %xmm0, %xmm14
+	vpunpckhqdq	%xmm4, %xmm2, %xmm0
+	vpshufd		$57, %xmm14, %xmm14
+	vpblendw	$51, %xmm15, %xmm0, %xmm15
+	vpaddd		%xmm15, %xmm11, %xmm15
+	vpaddd		%xmm14, %xmm15, %xmm15
+	vpxor		%xmm15, %xmm1, %xmm1
+	vpshufb		%xmm7, %xmm1, %xmm1
+	vpaddd		%xmm1, %xmm13, %xmm13
+	vpxor		%xmm13, %xmm14, %xmm14
+	vpslld		$20, %xmm14, %xmm11
+	vpsrld		$12, %xmm14, %xmm14
+	vpxor		%xmm11, %xmm14, %xmm14
+	vpblendw	$3, %xmm2, %xmm4, %xmm11
+	vpslldq		$8, %xmm11, %xmm0
+	vpblendw	$15, %xmm5, %xmm0, %xmm0
+	vpshufd		$99, %xmm0, %xmm0
+	vpaddd		%xmm15, %xmm0, %xmm15
+	vpaddd		%xmm14, %xmm15, %xmm15
+	vpxor		%xmm15, %xmm1, %xmm0
+	vpaddd		%xmm12, %xmm15, %xmm15
+	vpshufb		%xmm6, %xmm0, %xmm0
+	vpaddd		%xmm0, %xmm13, %xmm13
+	vpxor		%xmm13, %xmm14, %xmm14
+	vpshufd		$57, %xmm0, %xmm0
+	vpshufd		$78, %xmm13, %xmm13
+	vpslld		$25, %xmm14, %xmm1
+	vpsrld		$7, %xmm14, %xmm14
+	vpxor		%xmm1, %xmm14, %xmm14
+	vpblendw	$3, %xmm5, %xmm4, %xmm1
+	vpshufd		$147, %xmm14, %xmm14
+	vpaddd		%xmm14, %xmm15, %xmm15
+	vpxor		%xmm15, %xmm0, %xmm0
+	vpshufb		%xmm7, %xmm0, %xmm0
+	vpaddd		%xmm0, %xmm13, %xmm13
+	vpxor		%xmm13, %xmm14, %xmm14
+	vpslld		$20, %xmm14, %xmm12
+	vpsrld		$12, %xmm14, %xmm14
+	vpxor		%xmm12, %xmm14, %xmm14
+	vpsrldq		$4, %xmm2, %xmm12
+	vpblendw	$60, %xmm12, %xmm1, %xmm1
+	vpaddd		%xmm1, %xmm15, %xmm15
+	vpaddd		%xmm14, %xmm15, %xmm15
+	vpxor		%xmm15, %xmm0, %xmm0
+	vpblendw	$12, %xmm4, %xmm3, %xmm1
+	vpshufb		%xmm6, %xmm0, %xmm0
+	vpaddd		%xmm0, %xmm13, %xmm13
+	vpxor		%xmm13, %xmm14, %xmm14
+	vpshufd		$147, %xmm0, %xmm0
+	vpshufd		$78, %xmm13, %xmm13
+	vpslld		$25, %xmm14, %xmm12
+	vpsrld		$7, %xmm14, %xmm14
+	vpxor		%xmm12, %xmm14, %xmm14
+	vpsrldq		$4, %xmm5, %xmm12
+	vpblendw	$48, %xmm12, %xmm1, %xmm1
+	vpshufd		$33, %xmm5, %xmm12
+	vpshufd		$57, %xmm14, %xmm14
+	vpshufd		$108, %xmm1, %xmm1
+	vpblendw	$51, %xmm12, %xmm10, %xmm12
+	vpaddd		%xmm15, %xmm1, %xmm15
+	vpaddd		%xmm14, %xmm15, %xmm15
+	vpxor		%xmm15, %xmm0, %xmm0
+	vpaddd		%xmm12, %xmm15, %xmm15
+	vpshufb		%xmm7, %xmm0, %xmm0
+	vpaddd		%xmm0, %xmm13, %xmm1
+	vpxor		%xmm1, %xmm14, %xmm14
+	vpslld		$20, %xmm14, %xmm13
+	vpsrld		$12, %xmm14, %xmm14
+	vpxor		%xmm13, %xmm14, %xmm14
+	vpslldq		$12, %xmm3, %xmm13
+	vpaddd		%xmm14, %xmm15, %xmm15
+	vpxor		%xmm15, %xmm0, %xmm0
+	vpshufb		%xmm6, %xmm0, %xmm0
+	vpaddd		%xmm0, %xmm1, %xmm1
+	vpxor		%xmm1, %xmm14, %xmm14
+	vpshufd		$57, %xmm0, %xmm0
+	vpshufd		$78, %xmm1, %xmm1
+	vpslld		$25, %xmm14, %xmm12
+	vpsrld		$7, %xmm14, %xmm14
+	vpxor		%xmm12, %xmm14, %xmm14
+	vpblendw	$51, %xmm5, %xmm4, %xmm12
+	vpshufd		$147, %xmm14, %xmm14
+	vpblendw	$192, %xmm13, %xmm12, %xmm12
+	vpaddd		%xmm12, %xmm15, %xmm15
+	vpaddd		%xmm14, %xmm15, %xmm15
+	vpxor		%xmm15, %xmm0, %xmm0
+	vpsrldq		$4, %xmm3, %xmm12
+	vpshufb		%xmm7, %xmm0, %xmm0
+	vpaddd		%xmm0, %xmm1, %xmm1
+	vpxor		%xmm1, %xmm14, %xmm14
+	vpslld		$20, %xmm14, %xmm13
+	vpsrld		$12, %xmm14, %xmm14
+	vpxor		%xmm13, %xmm14, %xmm14
+	vpblendw	$48, %xmm2, %xmm5, %xmm13
+	vpblendw	$3, %xmm12, %xmm13, %xmm13
+	vpshufd		$156, %xmm13, %xmm13
+	vpaddd		%xmm15, %xmm13, %xmm15
+	vpaddd		%xmm14, %xmm15, %xmm15
+	vpxor		%xmm15, %xmm0, %xmm0
+	vpshufb		%xmm6, %xmm0, %xmm0
+	vpaddd		%xmm0, %xmm1, %xmm1
+	vpxor		%xmm1, %xmm14, %xmm14
+	vpshufd		$147, %xmm0, %xmm0
+	vpshufd		$78, %xmm1, %xmm1
+	vpslld		$25, %xmm14, %xmm13
+	vpsrld		$7, %xmm14, %xmm14
+	vpxor		%xmm13, %xmm14, %xmm14
+	vpunpcklqdq	%xmm2, %xmm4, %xmm13
+	vpshufd		$57, %xmm14, %xmm14
+	vpblendw	$12, %xmm12, %xmm13, %xmm12
+	vpshufd		$180, %xmm12, %xmm12
+	vpaddd		%xmm15, %xmm12, %xmm15
+	vpaddd		%xmm14, %xmm15, %xmm15
+	vpxor		%xmm15, %xmm0, %xmm0
+	vpshufb		%xmm7, %xmm0, %xmm0
+	vpaddd		%xmm0, %xmm1, %xmm1
+	vpxor		%xmm1, %xmm14, %xmm14
+	vpslld		$20, %xmm14, %xmm12
+	vpsrld		$12, %xmm14, %xmm14
+	vpxor		%xmm12, %xmm14, %xmm14
+	vpunpckhqdq	%xmm9, %xmm4, %xmm12
+	vpshufd		$198, %xmm12, %xmm12
+	vpaddd		%xmm15, %xmm12, %xmm15
+	vpaddd		%xmm14, %xmm15, %xmm15
+	vpxor		%xmm15, %xmm0, %xmm0
+	vpaddd		%xmm15, %xmm8, %xmm15
+	vpshufb		%xmm6, %xmm0, %xmm0
+	vpaddd		%xmm0, %xmm1, %xmm1
+	vpxor		%xmm1, %xmm14, %xmm14
+	vpshufd		$57, %xmm0, %xmm0
+	vpshufd		$78, %xmm1, %xmm1
+	vpslld		$25, %xmm14, %xmm12
+	vpsrld		$7, %xmm14, %xmm14
+	vpxor		%xmm12, %xmm14, %xmm14
+	vpsrldq		$4, %xmm4, %xmm12
+	vpshufd		$147, %xmm14, %xmm14
+	vpaddd		%xmm14, %xmm15, %xmm15
+	vpxor		%xmm15, %xmm0, %xmm0
+	vpshufb		%xmm7, %xmm0, %xmm0
+	vpaddd		%xmm0, %xmm1, %xmm1
+	vpxor		%xmm1, %xmm14, %xmm14
+	vpslld		$20, %xmm14, %xmm8
+	vpsrld		$12, %xmm14, %xmm14
+	vpxor		%xmm14, %xmm8, %xmm14
+	vpblendw	$48, %xmm5, %xmm2, %xmm8
+	vpblendw	$3, %xmm12, %xmm8, %xmm8
+	vpunpckhqdq	%xmm5, %xmm4, %xmm12
+	vpshufd		$75, %xmm8, %xmm8
+	vpblendw	$60, %xmm10, %xmm12, %xmm10
+	vpaddd		%xmm15, %xmm8, %xmm15
+	vpaddd		%xmm14, %xmm15, %xmm15
+	vpxor		%xmm0, %xmm15, %xmm0
+	vpshufd		$45, %xmm10, %xmm10
+	vpshufb		%xmm6, %xmm0, %xmm0
+	vpaddd		%xmm15, %xmm10, %xmm15
+	vpaddd		%xmm0, %xmm1, %xmm1
+	vpxor		%xmm1, %xmm14, %xmm14
+	vpshufd		$147, %xmm0, %xmm0
+	vpshufd		$78, %xmm1, %xmm1
+	vpslld		$25, %xmm14, %xmm8
+	vpsrld		$7, %xmm14, %xmm14
+	vpxor		%xmm14, %xmm8, %xmm8
+	vpshufd		$57, %xmm8, %xmm8
+	vpaddd		%xmm8, %xmm15, %xmm15
+	vpxor		%xmm0, %xmm15, %xmm0
+	vpshufb		%xmm7, %xmm0, %xmm0
+	vpaddd		%xmm0, %xmm1, %xmm1
+	vpxor		%xmm8, %xmm1, %xmm8
+	vpslld		$20, %xmm8, %xmm10
+	vpsrld		$12, %xmm8, %xmm8
+	vpxor		%xmm8, %xmm10, %xmm10
+	vpunpckldq	%xmm3, %xmm4, %xmm8
+	vpunpcklqdq	%xmm9, %xmm8, %xmm9
+	vpaddd		%xmm9, %xmm15, %xmm9
+	vpaddd		%xmm10, %xmm9, %xmm9
+	vpxor		%xmm0, %xmm9, %xmm8
+	vpshufb		%xmm6, %xmm8, %xmm8
+	vpaddd		%xmm8, %xmm1, %xmm1
+	vpxor		%xmm1, %xmm10, %xmm10
+	vpshufd		$57, %xmm8, %xmm8
+	vpshufd		$78, %xmm1, %xmm1
+	vpslld		$25, %xmm10, %xmm12
+	vpsrld		$7, %xmm10, %xmm10
+	vpxor		%xmm10, %xmm12, %xmm10
+	vpblendw	$48, %xmm4, %xmm3, %xmm12
+	vpshufd		$147, %xmm10, %xmm0
+	vpunpckhdq	%xmm5, %xmm3, %xmm10
+	vpshufd		$78, %xmm12, %xmm12
+	vpunpcklqdq	%xmm4, %xmm10, %xmm10
+	vpblendw	$192, %xmm2, %xmm10, %xmm10
+	vpshufhw	$78, %xmm10, %xmm10
+	vpaddd		%xmm10, %xmm9, %xmm10
+	vpaddd		%xmm0, %xmm10, %xmm10
+	vpxor		%xmm8, %xmm10, %xmm8
+	vpshufb		%xmm7, %xmm8, %xmm8
+	vpaddd		%xmm8, %xmm1, %xmm1
+	vpxor		%xmm0, %xmm1, %xmm9
+	vpslld		$20, %xmm9, %xmm0
+	vpsrld		$12, %xmm9, %xmm9
+	vpxor		%xmm9, %xmm0, %xmm0
+	vpunpckhdq	%xmm5, %xmm4, %xmm9
+	vpblendw	$240, %xmm9, %xmm2, %xmm13
+	vpshufd		$39, %xmm13, %xmm13
+	vpaddd		%xmm10, %xmm13, %xmm10
+	vpaddd		%xmm0, %xmm10, %xmm10
+	vpxor		%xmm8, %xmm10, %xmm8
+	vpblendw	$12, %xmm4, %xmm2, %xmm13
+	vpshufb		%xmm6, %xmm8, %xmm8
+	vpslldq		$4, %xmm13, %xmm13
+	vpblendw	$15, %xmm5, %xmm13, %xmm13
+	vpaddd		%xmm8, %xmm1, %xmm1
+	vpxor		%xmm1, %xmm0, %xmm0
+	vpaddd		%xmm13, %xmm10, %xmm13
+	vpshufd		$147, %xmm8, %xmm8
+	vpshufd		$78, %xmm1, %xmm1
+	vpslld		$25, %xmm0, %xmm14
+	vpsrld		$7, %xmm0, %xmm0
+	vpxor		%xmm0, %xmm14, %xmm14
+	vpshufd		$57, %xmm14, %xmm14
+	vpaddd		%xmm14, %xmm13, %xmm13
+	vpxor		%xmm8, %xmm13, %xmm8
+	vpaddd		%xmm13, %xmm12, %xmm12
+	vpshufb		%xmm7, %xmm8, %xmm8
+	vpaddd		%xmm8, %xmm1, %xmm1
+	vpxor		%xmm14, %xmm1, %xmm14
+	vpslld		$20, %xmm14, %xmm10
+	vpsrld		$12, %xmm14, %xmm14
+	vpxor		%xmm14, %xmm10, %xmm10
+	vpaddd		%xmm10, %xmm12, %xmm12
+	vpxor		%xmm8, %xmm12, %xmm8
+	vpshufb		%xmm6, %xmm8, %xmm8
+	vpaddd		%xmm8, %xmm1, %xmm1
+	vpxor		%xmm1, %xmm10, %xmm0
+	vpshufd		$57, %xmm8, %xmm8
+	vpshufd		$78, %xmm1, %xmm1
+	vpslld		$25, %xmm0, %xmm10
+	vpsrld		$7, %xmm0, %xmm0
+	vpxor		%xmm0, %xmm10, %xmm10
+	vpblendw	$48, %xmm2, %xmm3, %xmm0
+	vpblendw	$15, %xmm11, %xmm0, %xmm0
+	vpshufd		$147, %xmm10, %xmm10
+	vpshufd		$114, %xmm0, %xmm0
+	vpaddd		%xmm12, %xmm0, %xmm0
+	vpaddd		%xmm10, %xmm0, %xmm0
+	vpxor		%xmm8, %xmm0, %xmm8
+	vpshufb		%xmm7, %xmm8, %xmm8
+	vpaddd		%xmm8, %xmm1, %xmm1
+	vpxor		%xmm10, %xmm1, %xmm10
+	vpslld		$20, %xmm10, %xmm11
+	vpsrld		$12, %xmm10, %xmm10
+	vpxor		%xmm10, %xmm11, %xmm10
+	vpslldq		$4, %xmm4, %xmm11
+	vpblendw	$192, %xmm11, %xmm3, %xmm3
+	vpunpckldq	%xmm5, %xmm4, %xmm4
+	vpshufd		$99, %xmm3, %xmm3
+	vpaddd		%xmm0, %xmm3, %xmm3
+	vpaddd		%xmm10, %xmm3, %xmm3
+	vpxor		%xmm8, %xmm3, %xmm11
+	vpunpckldq	%xmm5, %xmm2, %xmm0
+	vpblendw	$192, %xmm2, %xmm5, %xmm2
+	vpshufb		%xmm6, %xmm11, %xmm11
+	vpunpckhqdq	%xmm0, %xmm9, %xmm0
+	vpblendw	$15, %xmm4, %xmm2, %xmm4
+	vpaddd		%xmm11, %xmm1, %xmm1
+	vpxor		%xmm1, %xmm10, %xmm10
+	vpshufd		$147, %xmm11, %xmm11
+	vpshufd		$201, %xmm0, %xmm0
+	vpslld		$25, %xmm10, %xmm8
+	vpsrld		$7, %xmm10, %xmm10
+	vpxor		%xmm10, %xmm8, %xmm10
+	vpshufd		$78, %xmm1, %xmm1
+	vpaddd		%xmm3, %xmm0, %xmm0
+	vpshufd		$27, %xmm4, %xmm4
+	vpshufd		$57, %xmm10, %xmm10
+	vpaddd		%xmm10, %xmm0, %xmm0
+	vpxor		%xmm11, %xmm0, %xmm11
+	vpaddd		%xmm0, %xmm4, %xmm0
+	vpshufb		%xmm7, %xmm11, %xmm7
+	vpaddd		%xmm7, %xmm1, %xmm1
+	vpxor		%xmm10, %xmm1, %xmm10
+	vpslld		$20, %xmm10, %xmm8
+	vpsrld		$12, %xmm10, %xmm10
+	vpxor		%xmm10, %xmm8, %xmm8
+	vpaddd		%xmm8, %xmm0, %xmm0
+	vpxor		%xmm7, %xmm0, %xmm7
+	vpshufb		%xmm6, %xmm7, %xmm6
+	vpaddd		%xmm6, %xmm1, %xmm1
+	vpxor		%xmm1, %xmm8, %xmm8
+	vpshufd		$78, %xmm1, %xmm1
+	vpshufd		$57, %xmm6, %xmm6
+	vpslld		$25, %xmm8, %xmm2
+	vpsrld		$7, %xmm8, %xmm8
+	vpxor		%xmm8, %xmm2, %xmm8
+	vpxor		(%rdi), %xmm1, %xmm1
+	vpshufd		$147, %xmm8, %xmm8
+	vpxor		%xmm0, %xmm1, %xmm0
+	vmovups		%xmm0, (%rdi)
+	vpxor		16(%rdi), %xmm8, %xmm0
+	vpxor		%xmm6, %xmm0, %xmm6
+	vmovups		%xmm6, 16(%rdi)
+	addq		$64, %rsi
+	decq		%rdx
+	jnz .Lbeginofloop
+.Lendofloop:
+	ret
+ENDPROC(blake2s_compress_avx)
+#endif /* CONFIG_AS_AVX */
+
+#ifdef CONFIG_AS_AVX512
+ENTRY(blake2s_compress_avx512_ymm)
+	vmovdqu		(%rdi),%xmm0
+	vmovdqu		0x10(%rdi),%xmm1
+	vmovdqu		0x20(%rdi),%xmm4
+	vmovq		%rcx,%xmm5
+	vmovdqa		IV(%rip),%xmm14
+	vmovdqa		IV+16(%rip),%xmm15
+	jmp		.Lblake2s_compress_avx512_ymm_mainloop
+.align 32
+.Lblake2s_compress_avx512_ymm_mainloop:
+	vmovdqa		%xmm0,%xmm10
+	vmovdqa		%xmm1,%xmm11
+	vpaddq		%xmm5,%xmm4,%xmm4
+	vmovdqa		%xmm14,%xmm2
+	vpxor		%xmm15,%xmm4,%xmm3
+	vmovdqu		(%rsi),%ymm6
+	vmovdqu		0x20(%rsi),%ymm7
+	addq		$0x40,%rsi
+	leaq		SIGMA(%rip),%rax
+	movb		$0xa,%cl
+.Lblake2s_compress_avx512_ymm_roundloop:
+	addq		$0x40,%rax
+	vmovdqa		-0x40(%rax),%ymm8
+	vmovdqa		-0x20(%rax),%ymm9
+	vpermi2d	%ymm7,%ymm6,%ymm8
+	vpermi2d	%ymm7,%ymm6,%ymm9
+	vmovdqa		%ymm8,%ymm6
+	vmovdqa		%ymm9,%ymm7
+	vpaddd		%xmm8,%xmm0,%xmm0
+	vpaddd		%xmm1,%xmm0,%xmm0
+	vpxor		%xmm0,%xmm3,%xmm3
+	vprord		$0x10,%xmm3,%xmm3
+	vpaddd		%xmm3,%xmm2,%xmm2
+	vpxor		%xmm2,%xmm1,%xmm1
+	vprord		$0xc,%xmm1,%xmm1
+	vextracti128	$0x1,%ymm8,%xmm8
+	vpaddd		%xmm8,%xmm0,%xmm0
+	vpaddd		%xmm1,%xmm0,%xmm0
+	vpxor		%xmm0,%xmm3,%xmm3
+	vprord		$0x8,%xmm3,%xmm3
+	vpaddd		%xmm3,%xmm2,%xmm2
+	vpxor		%xmm2,%xmm1,%xmm1
+	vprord		$0x7,%xmm1,%xmm1
+	vpshufd		$0x39,%xmm1,%xmm1
+	vpshufd		$0x4e,%xmm2,%xmm2
+	vpshufd		$0x93,%xmm3,%xmm3
+	vpaddd		%xmm9,%xmm0,%xmm0
+	vpaddd		%xmm1,%xmm0,%xmm0
+	vpxor		%xmm0,%xmm3,%xmm3
+	vprord		$0x10,%xmm3,%xmm3
+	vpaddd		%xmm3,%xmm2,%xmm2
+	vpxor		%xmm2,%xmm1,%xmm1
+	vprord		$0xc,%xmm1,%xmm1
+	vextracti128	$0x1,%ymm9,%xmm9
+	vpaddd		%xmm9,%xmm0,%xmm0
+	vpaddd		%xmm1,%xmm0,%xmm0
+	vpxor		%xmm0,%xmm3,%xmm3
+	vprord		$0x8,%xmm3,%xmm3
+	vpaddd		%xmm3,%xmm2,%xmm2
+	vpxor		%xmm2,%xmm1,%xmm1
+	vprord		$0x7,%xmm1,%xmm1
+	vpshufd		$0x93,%xmm1,%xmm1
+	vpshufd		$0x4e,%xmm2,%xmm2
+	vpshufd		$0x39,%xmm3,%xmm3
+	decb		%cl
+	jne		.Lblake2s_compress_avx512_ymm_roundloop
+	vpxor		%xmm10,%xmm0,%xmm0
+	vpxor		%xmm11,%xmm1,%xmm1
+	vpxor		%xmm2,%xmm0,%xmm0
+	vpxor		%xmm3,%xmm1,%xmm1
+	decq		%rdx
+	jne		.Lblake2s_compress_avx512_ymm_mainloop
+	vmovdqu		%xmm0,(%rdi)
+	vmovdqu		%xmm1,0x10(%rdi)
+	vmovdqu		%xmm4,0x20(%rdi)
+	vzeroupper
+	retq
+ENDPROC(blake2s_compress_avx512_ymm)
+
+ENTRY(blake2s_compress_avx512_zmm)
+	vmovdqu (%rdi),%xmm0
+	vmovdqu 0x10(%rdi),%xmm1
+	vmovdqu 0x20(%rdi),%xmm15
+	vmovq  %rcx,%xmm13
+	jmp .Lblake2s_compress_avx512_zmm_mainloop
+.align 32
+.Lblake2s_compress_avx512_zmm_mainloop:
+	vpaddq %xmm13,%xmm15,%xmm15
+	vmovdqa IV(%rip),%xmm2
+	vpxor  IV+16(%rip),%xmm15,%xmm3
+	lea    SIGMA(%rip),%rax
+	movl  $10,%ecx
+.Lblake2s_compress_avx512_zmm_roundloop:
+	add    $0x40,%rax
+	vmovdqa -0x40(%rax),%xmm7
+	vpcmpeqd %xmm14,%xmm14,%xmm14
+	vpgatherdd %xmm14,(%rsi,%xmm7,4),%xmm6
+	vpaddd %xmm6,%xmm0,%xmm0
+	vpaddd %xmm1,%xmm0,%xmm0
+	vpxor  %xmm0,%xmm3,%xmm3
+	vprord $0x10,%xmm3,%xmm3
+	vpaddd %xmm3,%xmm2,%xmm2
+	vpxor  %xmm2,%xmm1,%xmm1
+	vprord $0xc,%xmm1,%xmm1
+	vmovdqa -0x30(%rax),%xmm7
+	vpcmpeqd %xmm14,%xmm14,%xmm14
+	vpgatherdd %xmm14,(%rsi,%xmm7,4),%xmm6
+	vpaddd %xmm6,%xmm0,%xmm0
+	vpaddd %xmm1,%xmm0,%xmm0
+	vpxor  %xmm0,%xmm3,%xmm3
+	vprord $0x8,%xmm3,%xmm3
+	vpaddd %xmm3,%xmm2,%xmm2
+	vpxor  %xmm2,%xmm1,%xmm1
+	vprord $0x7,%xmm1,%xmm1
+	vpshufd $0x39,%xmm1,%xmm1
+	vpshufd $0x4e,%xmm2,%xmm2
+	vpshufd $0x93,%xmm3,%xmm3
+	vmovdqa -0x20(%rax),%xmm7
+	vpcmpeqd %xmm14,%xmm14,%xmm14
+	vpgatherdd %xmm14,(%rsi,%xmm7,4),%xmm6
+	vpaddd %xmm6,%xmm0,%xmm0
+	vpaddd %xmm1,%xmm0,%xmm0
+	vpxor  %xmm0,%xmm3,%xmm3
+	vprord $0x10,%xmm3,%xmm3
+	vpaddd %xmm3,%xmm2,%xmm2
+	vpxor  %xmm2,%xmm1,%xmm1
+	vprord $0xc,%xmm1,%xmm1
+	vmovdqa -0x10(%rax),%xmm7
+	vpcmpeqd %xmm14,%xmm14,%xmm14
+	vpgatherdd %xmm14,(%rsi,%xmm7,4),%xmm6
+	vpaddd %xmm6,%xmm0,%xmm0
+	vpaddd %xmm1,%xmm0,%xmm0
+	vpxor  %xmm0,%xmm3,%xmm3
+	vprord $0x8,%xmm3,%xmm3
+	vpaddd %xmm3,%xmm2,%xmm2
+	vpxor  %xmm2,%xmm1,%xmm1
+	vprord $0x7,%xmm1,%xmm1
+	vpshufd $0x93,%xmm1,%xmm1
+	vpshufd $0x4e,%xmm2,%xmm2
+	vpshufd $0x39,%xmm3,%xmm3
+	decl  %ecx
+	jne    .Lblake2s_compress_avx512_zmm_roundloop
+	add    $0x40,%rsi
+	vpxor  (%rdi),%xmm0,%xmm0
+	vpxor  0x10(%rdi),%xmm1,%xmm1
+	vpxor  %xmm2,%xmm0,%xmm0
+	vpxor  %xmm3,%xmm1,%xmm1
+	vmovdqu %xmm0,(%rdi)
+	vmovdqu %xmm1,0x10(%rdi)
+	dec    %rdx
+	jne    .Lblake2s_compress_avx512_zmm_mainloop
+	vmovdqu %xmm15,0x20(%rdi)
+	vzeroupper
+	retq
+ENDPROC(blake2s_compress_avx512_zmm)
+#endif /* CONFIG_AS_AVX512 */
diff --git a/curve25519-u128.c b/curve25519-u128.c
deleted file mode 100644
index b51d18a..0000000
--- a/curve25519-u128.c
+++ /dev/null
@@ -1,420 +0,0 @@
-/* SPDX-License-Identifier: GPL-2.0
- *
- * Copyright (C) 2008 Google Inc. All Rights Reserved.
- * Copyright (C) 2015-2018 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved.
- *
- * Original author: Adam Langley <agl@imperialviolet.org>
- */
-
-#include <linux/kernel.h>
-#include <linux/string.h>
-
-enum { CURVE25519_POINT_SIZE = 32 };
-
-typedef u64 limb;
-typedef limb felem[5];
-typedef __uint128_t u128;
-
-static __always_inline void normalize_secret(u8 secret[CURVE25519_POINT_SIZE])
-{
-	secret[0] &= 248;
-	secret[31] &= 127;
-	secret[31] |= 64;
-}
-
-/* Sum two numbers: output += in */
-static __always_inline void fsum(limb *output, const limb *in)
-{
-	output[0] += in[0];
-	output[1] += in[1];
-	output[2] += in[2];
-	output[3] += in[3];
-	output[4] += in[4];
-}
-
-/* Find the difference of two numbers: output = in - output
- * (note the order of the arguments!)
- *
- * Assumes that out[i] < 2**52
- * On return, out[i] < 2**55
- */
-static __always_inline void fdifference_backwards(felem out, const felem in)
-{
-	/* 152 is 19 << 3 */
-	static const limb two54m152 = (((limb)1) << 54) - 152;
-	static const limb two54m8 = (((limb)1) << 54) - 8;
-
-	out[0] = in[0] + two54m152 - out[0];
-	out[1] = in[1] + two54m8 - out[1];
-	out[2] = in[2] + two54m8 - out[2];
-	out[3] = in[3] + two54m8 - out[3];
-	out[4] = in[4] + two54m8 - out[4];
-}
-
-/* Multiply a number by a scalar: output = in * scalar */
-static __always_inline void fscalar_product(felem output, const felem in, const limb scalar)
-{
-	u128 a;
-
-	a = ((u128) in[0]) * scalar;
-	output[0] = ((limb)a) & 0x7ffffffffffffUL;
-
-	a = ((u128) in[1]) * scalar + ((limb) (a >> 51));
-	output[1] = ((limb)a) & 0x7ffffffffffffUL;
-
-	a = ((u128) in[2]) * scalar + ((limb) (a >> 51));
-	output[2] = ((limb)a) & 0x7ffffffffffffUL;
-
-	a = ((u128) in[3]) * scalar + ((limb) (a >> 51));
-	output[3] = ((limb)a) & 0x7ffffffffffffUL;
-
-	a = ((u128) in[4]) * scalar + ((limb) (a >> 51));
-	output[4] = ((limb)a) & 0x7ffffffffffffUL;
-
-	output[0] += (a >> 51) * 19;
-}
-
-/* Multiply two numbers: output = in2 * in
- *
- * output must be distinct to both inputs. The inputs are reduced coefficient
- * form, the output is not.
- *
- * Assumes that in[i] < 2**55 and likewise for in2.
- * On return, output[i] < 2**52
- */
-static __always_inline void fmul(felem output, const felem in2, const felem in)
-{
-	u128 t[5];
-	limb r0, r1, r2, r3, r4, s0, s1, s2, s3, s4, c;
-
-	r0 = in[0];
-	r1 = in[1];
-	r2 = in[2];
-	r3 = in[3];
-	r4 = in[4];
-
-	s0 = in2[0];
-	s1 = in2[1];
-	s2 = in2[2];
-	s3 = in2[3];
-	s4 = in2[4];
-
-	t[0]  =  ((u128) r0) * s0;
-	t[1]  =  ((u128) r0) * s1 + ((u128) r1) * s0;
-	t[2]  =  ((u128) r0) * s2 + ((u128) r2) * s0 + ((u128) r1) * s1;
-	t[3]  =  ((u128) r0) * s3 + ((u128) r3) * s0 + ((u128) r1) * s2 + ((u128) r2) * s1;
-	t[4]  =  ((u128) r0) * s4 + ((u128) r4) * s0 + ((u128) r3) * s1 + ((u128) r1) * s3 + ((u128) r2) * s2;
-
-	r4 *= 19;
-	r1 *= 19;
-	r2 *= 19;
-	r3 *= 19;
-
-	t[0] += ((u128) r4) * s1 + ((u128) r1) * s4 + ((u128) r2) * s3 + ((u128) r3) * s2;
-	t[1] += ((u128) r4) * s2 + ((u128) r2) * s4 + ((u128) r3) * s3;
-	t[2] += ((u128) r4) * s3 + ((u128) r3) * s4;
-	t[3] += ((u128) r4) * s4;
-
-			r0 = (limb)t[0] & 0x7ffffffffffffUL; c = (limb)(t[0] >> 51);
-	t[1] += c;      r1 = (limb)t[1] & 0x7ffffffffffffUL; c = (limb)(t[1] >> 51);
-	t[2] += c;      r2 = (limb)t[2] & 0x7ffffffffffffUL; c = (limb)(t[2] >> 51);
-	t[3] += c;      r3 = (limb)t[3] & 0x7ffffffffffffUL; c = (limb)(t[3] >> 51);
-	t[4] += c;      r4 = (limb)t[4] & 0x7ffffffffffffUL; c = (limb)(t[4] >> 51);
-	r0 +=   c * 19; c = r0 >> 51; r0 = r0 & 0x7ffffffffffffUL;
-	r1 +=   c;      c = r1 >> 51; r1 = r1 & 0x7ffffffffffffUL;
-	r2 +=   c;
-
-	output[0] = r0;
-	output[1] = r1;
-	output[2] = r2;
-	output[3] = r3;
-	output[4] = r4;
-}
-
-static __always_inline void fsquare_times(felem output, const felem in, limb count)
-{
-	u128 t[5];
-	limb r0, r1, r2, r3, r4, c;
-	limb d0, d1, d2, d4, d419;
-
-	r0 = in[0];
-	r1 = in[1];
-	r2 = in[2];
-	r3 = in[3];
-	r4 = in[4];
-
-	do {
-		d0 = r0 * 2;
-		d1 = r1 * 2;
-		d2 = r2 * 2 * 19;
-		d419 = r4 * 19;
-		d4 = d419 * 2;
-
-		t[0] = ((u128) r0) * r0 + ((u128) d4) * r1 + (((u128) d2) * (r3     ));
-		t[1] = ((u128) d0) * r1 + ((u128) d4) * r2 + (((u128) r3) * (r3 * 19));
-		t[2] = ((u128) d0) * r2 + ((u128) r1) * r1 + (((u128) d4) * (r3     ));
-		t[3] = ((u128) d0) * r3 + ((u128) d1) * r2 + (((u128) r4) * (d419   ));
-		t[4] = ((u128) d0) * r4 + ((u128) d1) * r3 + (((u128) r2) * (r2     ));
-
-				r0 = (limb)t[0] & 0x7ffffffffffffUL; c = (limb)(t[0] >> 51);
-		t[1] += c;      r1 = (limb)t[1] & 0x7ffffffffffffUL; c = (limb)(t[1] >> 51);
-		t[2] += c;      r2 = (limb)t[2] & 0x7ffffffffffffUL; c = (limb)(t[2] >> 51);
-		t[3] += c;      r3 = (limb)t[3] & 0x7ffffffffffffUL; c = (limb)(t[3] >> 51);
-		t[4] += c;      r4 = (limb)t[4] & 0x7ffffffffffffUL; c = (limb)(t[4] >> 51);
-		r0 +=   c * 19; c = r0 >> 51; r0 = r0 & 0x7ffffffffffffUL;
-		r1 +=   c;      c = r1 >> 51; r1 = r1 & 0x7ffffffffffffUL;
-		r2 +=   c;
-	} while (--count);
-
-	output[0] = r0;
-	output[1] = r1;
-	output[2] = r2;
-	output[3] = r3;
-	output[4] = r4;
-}
-
-/* Load a little-endian 64-bit number  */
-static inline limb load_limb(const u8 *in)
-{
-	return le64_to_cpu(*(__le64 *)in);
-}
-
-static inline void store_limb(u8 *out, limb in)
-{
-	*(__le64 *)out = cpu_to_le64(in);
-}
-
-/* Take a little-endian, 32-byte number and expand it into polynomial form */
-static inline void fexpand(limb *output, const u8 *in)
-{
-	output[0] = load_limb(in) & 0x7ffffffffffffUL;
-	output[1] = (load_limb(in + 6) >> 3) & 0x7ffffffffffffUL;
-	output[2] = (load_limb(in + 12) >> 6) & 0x7ffffffffffffUL;
-	output[3] = (load_limb(in + 19) >> 1) & 0x7ffffffffffffUL;
-	output[4] = (load_limb(in + 24) >> 12) & 0x7ffffffffffffUL;
-}
-
-/* Take a fully reduced polynomial form number and contract it into a
- * little-endian, 32-byte array
- */
-static void fcontract(u8 *output, const felem input)
-{
-	u128 t[5];
-
-	t[0] = input[0];
-	t[1] = input[1];
-	t[2] = input[2];
-	t[3] = input[3];
-	t[4] = input[4];
-
-	t[1] += t[0] >> 51; t[0] &= 0x7ffffffffffffUL;
-	t[2] += t[1] >> 51; t[1] &= 0x7ffffffffffffUL;
-	t[3] += t[2] >> 51; t[2] &= 0x7ffffffffffffUL;
-	t[4] += t[3] >> 51; t[3] &= 0x7ffffffffffffUL;
-	t[0] += 19 * (t[4] >> 51); t[4] &= 0x7ffffffffffffUL;
-
-	t[1] += t[0] >> 51; t[0] &= 0x7ffffffffffffUL;
-	t[2] += t[1] >> 51; t[1] &= 0x7ffffffffffffUL;
-	t[3] += t[2] >> 51; t[2] &= 0x7ffffffffffffUL;
-	t[4] += t[3] >> 51; t[3] &= 0x7ffffffffffffUL;
-	t[0] += 19 * (t[4] >> 51); t[4] &= 0x7ffffffffffffUL;
-
-	/* now t is between 0 and 2^255-1, properly carried. */
-	/* case 1: between 0 and 2^255-20. case 2: between 2^255-19 and 2^255-1. */
-
-	t[0] += 19;
-
-	t[1] += t[0] >> 51; t[0] &= 0x7ffffffffffffUL;
-	t[2] += t[1] >> 51; t[1] &= 0x7ffffffffffffUL;
-	t[3] += t[2] >> 51; t[2] &= 0x7ffffffffffffUL;
-	t[4] += t[3] >> 51; t[3] &= 0x7ffffffffffffUL;
-	t[0] += 19 * (t[4] >> 51); t[4] &= 0x7ffffffffffffUL;
-
-	/* now between 19 and 2^255-1 in both cases, and offset by 19. */
-
-	t[0] += 0x8000000000000UL - 19;
-	t[1] += 0x8000000000000UL - 1;
-	t[2] += 0x8000000000000UL - 1;
-	t[3] += 0x8000000000000UL - 1;
-	t[4] += 0x8000000000000UL - 1;
-
-	/* now between 2^255 and 2^256-20, and offset by 2^255. */
-
-	t[1] += t[0] >> 51; t[0] &= 0x7ffffffffffffUL;
-	t[2] += t[1] >> 51; t[1] &= 0x7ffffffffffffUL;
-	t[3] += t[2] >> 51; t[2] &= 0x7ffffffffffffUL;
-	t[4] += t[3] >> 51; t[3] &= 0x7ffffffffffffUL;
-	t[4] &= 0x7ffffffffffffUL;
-
-	store_limb(output,    t[0] | (t[1] << 51));
-	store_limb(output+8,  (t[1] >> 13) | (t[2] << 38));
-	store_limb(output+16, (t[2] >> 26) | (t[3] << 25));
-	store_limb(output+24, (t[3] >> 39) | (t[4] << 12));
-}
-
-/* Input: Q, Q', Q-Q'
- * Output: 2Q, Q+Q'
- *
- *   x2 z3: long form
- *   x3 z3: long form
- *   x z: short form, destroyed
- *   xprime zprime: short form, destroyed
- *   qmqp: short form, preserved
- */
-static void fmonty(limb *x2, limb *z2, /* output 2Q */
-			 limb *x3, limb *z3, /* output Q + Q' */
-			 limb *x, limb *z,   /* input Q */
-			 limb *xprime, limb *zprime, /* input Q' */
-
-			 const limb *qmqp /* input Q - Q' */)
-{
-	limb origx[5], origxprime[5], zzz[5], xx[5], zz[5], xxprime[5], zzprime[5], zzzprime[5];
-
-	memcpy(origx, x, 5 * sizeof(limb));
-	fsum(x, z);
-	fdifference_backwards(z, origx);  // does x - z
-
-	memcpy(origxprime, xprime, sizeof(limb) * 5);
-	fsum(xprime, zprime);
-	fdifference_backwards(zprime, origxprime);
-	fmul(xxprime, xprime, z);
-	fmul(zzprime, x, zprime);
-	memcpy(origxprime, xxprime, sizeof(limb) * 5);
-	fsum(xxprime, zzprime);
-	fdifference_backwards(zzprime, origxprime);
-	fsquare_times(x3, xxprime, 1);
-	fsquare_times(zzzprime, zzprime, 1);
-	fmul(z3, zzzprime, qmqp);
-
-	fsquare_times(xx, x, 1);
-	fsquare_times(zz, z, 1);
-	fmul(x2, xx, zz);
-	fdifference_backwards(zz, xx);  // does zz = xx - zz
-	fscalar_product(zzz, zz, 121665);
-	fsum(zzz, xx);
-	fmul(z2, zz, zzz);
-}
-
-/* Maybe swap the contents of two limb arrays (@a and @b), each @len elements
- * long. Perform the swap iff @swap is non-zero.
- *
- * This function performs the swap without leaking any side-channel
- * information.
- */
-static void swap_conditional(limb a[5], limb b[5], limb iswap)
-{
-	unsigned int i;
-	const limb swap = -iswap;
-
-	for (i = 0; i < 5; ++i) {
-		const limb x = swap & (a[i] ^ b[i]);
-
-		a[i] ^= x;
-		b[i] ^= x;
-	}
-}
-
-/* Calculates nQ where Q is the x-coordinate of a point on the curve
- *
- *   resultx/resultz: the x coordinate of the resulting curve point (short form)
- *   n: a little endian, 32-byte number
- *   q: a point of the curve (short form)
- */
-static void cmult(limb *resultx, limb *resultz, const u8 *n, const limb *q)
-{
-	limb a[5] = {0}, b[5] = {1}, c[5] = {1}, d[5] = {0};
-	limb *nqpqx = a, *nqpqz = b, *nqx = c, *nqz = d, *t;
-	limb e[5] = {0}, f[5] = {1}, g[5] = {0}, h[5] = {1};
-	limb *nqpqx2 = e, *nqpqz2 = f, *nqx2 = g, *nqz2 = h;
-
-	unsigned int i, j;
-
-	memcpy(nqpqx, q, sizeof(limb) * 5);
-
-	for (i = 0; i < 32; ++i) {
-		u8 byte = n[31 - i];
-
-		for (j = 0; j < 8; ++j) {
-			const limb bit = byte >> 7;
-
-			swap_conditional(nqx, nqpqx, bit);
-			swap_conditional(nqz, nqpqz, bit);
-			fmonty(nqx2, nqz2,
-						 nqpqx2, nqpqz2,
-						 nqx, nqz,
-						 nqpqx, nqpqz,
-						 q);
-			swap_conditional(nqx2, nqpqx2, bit);
-			swap_conditional(nqz2, nqpqz2, bit);
-
-			t = nqx;
-			nqx = nqx2;
-			nqx2 = t;
-			t = nqz;
-			nqz = nqz2;
-			nqz2 = t;
-			t = nqpqx;
-			nqpqx = nqpqx2;
-			nqpqx2 = t;
-			t = nqpqz;
-			nqpqz = nqpqz2;
-			nqpqz2 = t;
-
-			byte <<= 1;
-		}
-	}
-
-	memcpy(resultx, nqx, sizeof(limb) * 5);
-	memcpy(resultz, nqz, sizeof(limb) * 5);
-}
-
-static void crecip(felem out, const felem z)
-{
-	felem a, t0, b, c;
-
-	/* 2 */ fsquare_times(a, z, 1); // a = 2
-	/* 8 */ fsquare_times(t0, a, 2);
-	/* 9 */ fmul(b, t0, z); // b = 9
-	/* 11 */ fmul(a, b, a); // a = 11
-	/* 22 */ fsquare_times(t0, a, 1);
-	/* 2^5 - 2^0 = 31 */ fmul(b, t0, b);
-	/* 2^10 - 2^5 */ fsquare_times(t0, b, 5);
-	/* 2^10 - 2^0 */ fmul(b, t0, b);
-	/* 2^20 - 2^10 */ fsquare_times(t0, b, 10);
-	/* 2^20 - 2^0 */ fmul(c, t0, b);
-	/* 2^40 - 2^20 */ fsquare_times(t0, c, 20);
-	/* 2^40 - 2^0 */ fmul(t0, t0, c);
-	/* 2^50 - 2^10 */ fsquare_times(t0, t0, 10);
-	/* 2^50 - 2^0 */ fmul(b, t0, b);
-	/* 2^100 - 2^50 */ fsquare_times(t0, b, 50);
-	/* 2^100 - 2^0 */ fmul(c, t0, b);
-	/* 2^200 - 2^100 */ fsquare_times(t0, c, 100);
-	/* 2^200 - 2^0 */ fmul(t0, t0, c);
-	/* 2^250 - 2^50 */ fsquare_times(t0, t0, 50);
-	/* 2^250 - 2^0 */ fmul(t0, t0, b);
-	/* 2^255 - 2^5 */ fsquare_times(t0, t0, 5);
-	/* 2^255 - 21 */ fmul(out, t0, a);
-}
-
-bool curve25519(u8 mypublic[CURVE25519_POINT_SIZE], const u8 secret[CURVE25519_POINT_SIZE], const u8 basepoint[CURVE25519_POINT_SIZE])
-{
-	limb bp[5], x[5], z[5], zmone[5];
-	u8 e[32];
-
-	memcpy(e, secret, 32);
-	normalize_secret(e);
-
-	fexpand(bp, basepoint);
-	cmult(x, z, e, bp);
-	crecip(zmone, z);
-	fmul(z, x, zmone);
-	fcontract(mypublic, z);
-
-	memzero_explicit(e, sizeof(e));
-	memzero_explicit(bp, sizeof(bp));
-	memzero_explicit(x, sizeof(x));
-	memzero_explicit(z, sizeof(z));
-	memzero_explicit(zmone, sizeof(zmone));
-
-	return true;
-}
diff --git a/function.h b/function.h
deleted file mode 100644
index 6f360e9..0000000
--- a/function.h
+++ /dev/null
@@ -1,68 +0,0 @@
-/* SPDX-License-Identifier: GPL-2.0
- *
- * Copyright (C) 2018 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved.
- */
-
-enum { WARMUP = 5000, TRIALS = 10000, IDLE = 1 * 1000 };
-
-enum { CURVE25519_POINT_SIZE = 32 };
-bool curve25519(u8 mypublic[CURVE25519_POINT_SIZE], const u8 secret[CURVE25519_POINT_SIZE], const u8 basepoint[CURVE25519_POINT_SIZE]);
-
-struct curve25519_test_vector {
-	u8 private[CURVE25519_POINT_SIZE];
-	u8 public[CURVE25519_POINT_SIZE];
-	u8 result[CURVE25519_POINT_SIZE];
-	bool valid;
-};
-
-static const struct curve25519_test_vector curve25519_test_vectors[] __initconst = {
-	{
-		.private = { 0x77, 0x07, 0x6d, 0x0a, 0x73, 0x18, 0xa5, 0x7d, 0x3c, 0x16, 0xc1, 0x72, 0x51, 0xb2, 0x66, 0x45, 0xdf, 0x4c, 0x2f, 0x87, 0xeb, 0xc0, 0x99, 0x2a, 0xb1, 0x77, 0xfb, 0xa5, 0x1d, 0xb9, 0x2c, 0x2a },
-		.public = { 0xde, 0x9e, 0xdb, 0x7d, 0x7b, 0x7d, 0xc1, 0xb4, 0xd3, 0x5b, 0x61, 0xc2, 0xec, 0xe4, 0x35, 0x37, 0x3f, 0x83, 0x43, 0xc8, 0x5b, 0x78, 0x67, 0x4d, 0xad, 0xfc, 0x7e, 0x14, 0x6f, 0x88, 0x2b, 0x4f },
-		.result = { 0x4a, 0x5d, 0x9d, 0x5b, 0xa4, 0xce, 0x2d, 0xe1, 0x72, 0x8e, 0x3b, 0xf4, 0x80, 0x35, 0x0f, 0x25, 0xe0, 0x7e, 0x21, 0xc9, 0x47, 0xd1, 0x9e, 0x33, 0x76, 0xf0, 0x9b, 0x3c, 0x1e, 0x16, 0x17, 0x42 },
-		.valid = true
-	},
-	{
-		.private = { 0x5d, 0xab, 0x08, 0x7e, 0x62, 0x4a, 0x8a, 0x4b, 0x79, 0xe1, 0x7f, 0x8b, 0x83, 0x80, 0x0e, 0xe6, 0x6f, 0x3b, 0xb1, 0x29, 0x26, 0x18, 0xb6, 0xfd, 0x1c, 0x2f, 0x8b, 0x27, 0xff, 0x88, 0xe0, 0xeb },
-		.public = { 0x85, 0x20, 0xf0, 0x09, 0x89, 0x30, 0xa7, 0x54, 0x74, 0x8b, 0x7d, 0xdc, 0xb4, 0x3e, 0xf7, 0x5a, 0x0d, 0xbf, 0x3a, 0x0d, 0x26, 0x38, 0x1a, 0xf4, 0xeb, 0xa4, 0xa9, 0x8e, 0xaa, 0x9b, 0x4e, 0x6a },
-		.result = { 0x4a, 0x5d, 0x9d, 0x5b, 0xa4, 0xce, 0x2d, 0xe1, 0x72, 0x8e, 0x3b, 0xf4, 0x80, 0x35, 0x0f, 0x25, 0xe0, 0x7e, 0x21, 0xc9, 0x47, 0xd1, 0x9e, 0x33, 0x76, 0xf0, 0x9b, 0x3c, 0x1e, 0x16, 0x17, 0x42 },
-		.valid = true
-	},
-	{
-		.private = { 1 },
-		.public = { 0x25, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 },
-		.result = { 0x3c, 0x77, 0x77, 0xca, 0xf9, 0x97, 0xb2, 0x64, 0x41, 0x60, 0x77, 0x66, 0x5b, 0x4e, 0x22, 0x9d, 0xb, 0x95, 0x48, 0xdc, 0xc, 0xd8, 0x19, 0x98, 0xdd, 0xcd, 0xc5, 0xc8, 0x53, 0x3c, 0x79, 0x7f },
-		.valid = true
-	},
-	{
-		.private = { 1 },
-		.public = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff },
-		.result = { 0xb3, 0x2d, 0x13, 0x62, 0xc2, 0x48, 0xd6, 0x2f, 0xe6, 0x26, 0x19, 0xcf, 0xf0, 0x4d, 0xd4, 0x3d, 0xb7, 0x3f, 0xfc, 0x1b, 0x63, 0x8, 0xed, 0xe3, 0xb, 0x78, 0xd8, 0x73, 0x80, 0xf1, 0xe8, 0x34 },
-		.valid = true
-	},
-	{
-		.private = { 0xa5, 0x46, 0xe3, 0x6b, 0xf0, 0x52, 0x7c, 0x9d, 0x3b, 0x16, 0x15, 0x4b, 0x82, 0x46, 0x5e, 0xdd, 0x62, 0x14, 0x4c, 0x0a, 0xc1, 0xfc, 0x5a, 0x18, 0x50, 0x6a, 0x22, 0x44, 0xba, 0x44, 0x9a, 0xc4 },
-		.public = { 0xe6, 0xdb, 0x68, 0x67, 0x58, 0x30, 0x30, 0xdb, 0x35, 0x94, 0xc1, 0xa4, 0x24, 0xb1, 0x5f, 0x7c, 0x72, 0x66, 0x24, 0xec, 0x26, 0xb3, 0x35, 0x3b, 0x10, 0xa9, 0x03, 0xa6, 0xd0, 0xab, 0x1c, 0x4c },
-		.result = { 0xc3, 0xda, 0x55, 0x37, 0x9d, 0xe9, 0xc6, 0x90, 0x8e, 0x94, 0xea, 0x4d, 0xf2, 0x8d, 0x08, 0x4f, 0x32, 0xec, 0xcf, 0x03, 0x49, 0x1c, 0x71, 0xf7, 0x54, 0xb4, 0x07, 0x55, 0x77, 0xa2, 0x85, 0x52 },
-		.valid = true
-	},
-	{
-		.private = { 1, 2, 3, 4 },
-		.public = { 0 },
-		.result = { 0 },
-		.valid = false
-	},
-	{
-		.private = { 2, 4, 6, 8 },
-		.public = { 0xe0, 0xeb, 0x7a, 0x7c, 0x3b, 0x41, 0xb8, 0xae, 0x16, 0x56, 0xe3, 0xfa, 0xf1, 0x9f, 0xc4, 0x6a, 0xda, 0x09, 0x8d, 0xeb, 0x9c, 0x32, 0xb1, 0xfd, 0x86, 0x62, 0x05, 0x16, 0x5f, 0x49, 0xb8 },
-		.result = { 0 },
-		.valid = false
-	}
-};
-
-u8 dummy_out[CURVE25519_POINT_SIZE];
-
-static __always_inline int function(void)
-{
-	return curve25519(dummy_out, curve25519_test_vectors[0].private, curve25519_test_vectors[0].public);
-}
diff --git a/main.c b/main.c
index d538f48..ad8de4c 100644
--- a/main.c
+++ b/main.c
@@ -6,37 +6,95 @@
 #include <linux/init.h>
 #include <linux/module.h>
 #include <linux/delay.h>
-#include "function.h"
+#include <linux/slab.h>
+#include <linux/sort.h>
+#include <asm/cpufeature.h>
+#include <asm/processor.h>
+#include <asm/fpu/api.h>
+#include <asm/simd.h>
 
 static unsigned long stamp = 0;
 module_param(stamp, ulong, 0);
 int dummy;
 
+enum { BLOCKS_PER_CALL = 16 };
+
+static u8 state[128];
+static u8 input[64 * BLOCKS_PER_CALL];
+
+#define declare_it(name) \
+asmlinkage void blake2s_compress_ ## name(u8 state[128], const u8 *block, const size_t nblocks, const u32 inc); \
+static __always_inline u8 name(void) \
+{ \
+	blake2s_compress_ ## name(state, input, BLOCKS_PER_CALL, 0); \
+	return input[0]; \
+}
+
+#define do_it(name) do { \
+	u32 eax = 0, ebx = 0, ecx = 0, edx = 0; \
+	for (i = 0; i < WARMUP; ++i) \
+		ret |= name(); \
+	asm volatile("cpuid" : "+a" (eax), "=b" (ebx), "=d" (edx), "+c" (ecx)); \
+	for (i = 0; i <= TRIALS; ++i) { \
+		trial_times[i] = get_cycles(); \
+		ret |= name(); \
+	} \
+	for (i = 0; i < TRIALS; ++i) \
+		trial_times[i] = trial_times[i + 1] - trial_times[i]; \
+	sort(trial_times, TRIALS + 1, sizeof(cycles_t), compare_cycles, NULL); \
+	median_ ## name = trial_times[TRIALS / 2]; \
+} while (0)
+
+#define report_it(name) do { \
+	pr_err("%lu: %12s: %6llu cycles per block\n", stamp, #name, median_ ## name / BLOCKS_PER_CALL); \
+} while (0)
+
+
+declare_it(avx)
+declare_it(avx512_ymm)
+declare_it(avx512_zmm)
+
+static int compare_cycles(const void *a, const void *b)
+{
+	return *((cycles_t *)a) - *((cycles_t *)b);
+}
+
 static int __init mod_init(void)
 {
+	enum { WARMUP = 6000, TRIALS = 5000, IDLE = 1 * 1000 };
 	int ret = 0, i;
-	cycles_t start, end;
+	cycles_t *trial_times;
+	cycles_t median_avx = 0;
+	cycles_t median_avx512_ymm = 0;
+	cycles_t median_avx512_zmm = 0;
 	unsigned long flags;
 	DEFINE_SPINLOCK(lock);
-	
+
+	trial_times = kcalloc(TRIALS + 1, sizeof(cycles_t), GFP_KERNEL);
+	if (!trial_times)
+		return -ENOMEM;
+
 	msleep(IDLE);
 
 	spin_lock_irqsave(&lock, flags);
-	
-	for (i = 0; i < WARMUP; ++i)
-		ret |= function();
 
-	start = get_cycles();
-	for (i = 0; i < TRIALS; ++i)
-		ret |= function();
-	end = get_cycles();
+	kernel_fpu_begin();
+
+	do_it(avx);
+	do_it(avx512_ymm);
+	do_it(avx512_zmm);
+
+	kernel_fpu_end();
 
 	spin_unlock_irqrestore(&lock, flags);
 	
-	pr_err("%lu: %llu cycles per call\n", stamp, (end - start) / TRIALS);
+	report_it(avx);
+	report_it(avx512_ymm);
+	report_it(avx512_zmm);
 
 	/* Don't let compiler be too clever. */
 	dummy = ret;
+	kfree(trial_times);
 	
 	/* We should never actually agree to insert the module. Choosing
 	 * -0x1000 here is an amazing hack. It causes the kernel to not
diff --git a/run.sh b/run.sh
index dfe5fb9..39c4719 100755
--- a/run.sh
+++ b/run.sh
@@ -8,9 +8,23 @@ nob_cpus() {
 	done
 }
 
+noturbo() {
+	echo "[+] Setting no-turbo to status $1"
+	if [[ -e /sys/devices/system/cpu/intel_pstate/no_turbo ]]; then
+		echo "$1" > /sys/devices/system/cpu/intel_pstate/no_turbo
+	else
+		local val
+		[[ $1 == 0 ]] && val=0x850089
+		[[ $1 == 1 ]] && val=0x4000850089
+		[[ -n $val ]] || return 0
+		wrmsr -a 0x1a0 $val
+	fi
+}
+
 [[ -e kbench9000.ko ]]
 
-trap "nob_cpus 1" INT TERM EXIT
+trap "nob_cpus 1; noturbo 0;" INT TERM EXIT
+noturbo 1
 nob_cpus 0
 
 echo "[+] Inserting module to run tests"