crypto/internal/bigmod: provide assembly addMulVVW* for riscv64

This provides an assembly implementation of addMulVVW* for riscv64,
processing four words per loop, resulting in a performance gain
of 23%+ for RSA decryption/signing on a StarFive VisionFive 2:

                       │    rsa1     │                rsa2                 │
                       │   sec/op    │   sec/op     vs base                │
DecryptPKCS1v15/2048-4   24.29m ± 0%   18.65m ± 0%  -23.24% (p=0.000 n=10)
DecryptPKCS1v15/3072-4   73.28m ± 0%   54.08m ± 0%  -26.20% (p=0.000 n=10)
DecryptPKCS1v15/4096-4   163.5m ± 0%   119.1m ± 0%  -27.17% (p=0.000 n=10)
EncryptPKCS1v15/2048-4   1.505m ± 0%   1.446m ± 0%   -3.93% (p=0.000 n=10)
DecryptOAEP/2048-4       24.37m ± 0%   18.72m ± 0%  -23.17% (p=0.000 n=10)
EncryptOAEP/2048-4       1.570m ± 0%   1.510m ± 0%   -3.84% (p=0.000 n=10)
SignPKCS1v15/2048-4      24.52m ± 0%   18.80m ± 0%  -23.36% (p=0.000 n=10)
VerifyPKCS1v15/2048-4    1.491m ± 0%   1.431m ± 0%   -4.00% (p=0.000 n=10)
SignPSS/2048-4           24.60m ± 0%   18.89m ± 0%  -23.21% (p=0.000 n=10)
VerifyPSS/2048-4         1.565m ± 0%   1.504m ± 0%   -3.87% (p=0.000 n=10)
geomean                  10.90m        9.066m       -16.79%

Change-Id: I8414ba0028b0781a945610abe02c285d2387aef3
Reviewed-on: https://go-review.googlesource.com/c/go/+/516536
Reviewed-by: Mark Ryan <markdryan@rivosinc.com>
Reviewed-by: Filippo Valsorda <filippo@golang.org>
Reviewed-by: Dmitri Shuralyov <dmitshur@google.com>
Reviewed-by: M Zhuo <mzh@golangcn.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Run-TryBot: Joel Sing <joel@sing.id.au>
TryBot-Result: Gopher Robot <gobot@golang.org>

3 files changed, 93 insertions, 2 deletions

diff --git a/src/crypto/internal/bigmod/nat_asm.go b/src/crypto/internal/bigmod/nat_asm.go
index 5eb91e1c6c..0283b07e68 100644
--- a/src/crypto/internal/bigmod/nat_asm.go
+++ b/src/crypto/internal/bigmod/nat_asm.go
@@ -2,7 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-//go:build !purego && (386 || amd64 || arm || arm64 || ppc64 || ppc64le || s390x)
+//go:build !purego && (386 || amd64 || arm || arm64 || ppc64 || ppc64le || riscv64 || s390x)
 
 package bigmod
 
diff --git a/src/crypto/internal/bigmod/nat_noasm.go b/src/crypto/internal/bigmod/nat_noasm.go
index eff12536f9..71f38da754 100644
--- a/src/crypto/internal/bigmod/nat_noasm.go
+++ b/src/crypto/internal/bigmod/nat_noasm.go
@@ -2,7 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-//go:build purego || !(386 || amd64 || arm || arm64 || ppc64 || ppc64le || s390x)
+//go:build purego || !(386 || amd64 || arm || arm64 || ppc64 || ppc64le || riscv64 || s390x)
 
 package bigmod
 
diff --git a/src/crypto/internal/bigmod/nat_riscv64.s b/src/crypto/internal/bigmod/nat_riscv64.s
new file mode 100644
index 0000000000..1d8c8c8900
--- /dev/null
+++ b/src/crypto/internal/bigmod/nat_riscv64.s
@@ -0,0 +1,91 @@
+// Copyright 2023 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build !purego
+
+#include "textflag.h"
+
+// func addMulVVW1024(z, x *uint, y uint) (c uint)
+TEXT ·addMulVVW1024(SB),$0-32
+	MOV	$16, X30
+	JMP	addMulVVWx(SB)
+
+// func addMulVVW1536(z, x *uint, y uint) (c uint)
+TEXT ·addMulVVW1536(SB),$0-32
+	MOV	$24, X30
+	JMP	addMulVVWx(SB)
+
+// func addMulVVW2048(z, x *uint, y uint) (c uint)
+TEXT ·addMulVVW2048(SB),$0-32
+	MOV	$32, X30
+	JMP	addMulVVWx(SB)
+
+TEXT addMulVVWx(SB),NOFRAME|NOSPLIT,$0
+	MOV	z+0(FP), X5
+	MOV	x+8(FP), X7
+	MOV	y+16(FP), X6
+	MOV	$0, X29
+
+	BEQZ	X30, done
+loop:
+	MOV	0*8(X5), X10	// z[0]
+	MOV	1*8(X5), X13	// z[1]
+	MOV	2*8(X5), X16	// z[2]
+	MOV	3*8(X5), X19	// z[3]
+
+	MOV	0*8(X7), X8	// x[0]
+	MOV	1*8(X7), X11	// x[1]
+	MOV	2*8(X7), X14	// x[2]
+	MOV	3*8(X7), X17	// x[3]
+
+	MULHU	X8, X6, X9	// z_hi[0] = x[0] * y
+	MUL	X8, X6, X8	// z_lo[0] = x[0] * y
+	ADD	X8, X10, X21	// z_lo[0] = x[0] * y + z[0]
+	SLTU	X8, X21, X22
+	ADD	X9, X22, X9	// z_hi[0] = x[0] * y + z[0]
+	ADD	X21, X29, X10	// z_lo[0] = x[0] * y + z[0] + c
+	SLTU	X21, X10, X22
+	ADD	X9, X22, X29	// next c
+
+	MULHU	X11, X6, X12	// z_hi[1] = x[1] * y
+	MUL	X11, X6, X11	// z_lo[1] = x[1] * y
+	ADD	X11, X13, X21	// z_lo[1] = x[1] * y + z[1]
+	SLTU	X11, X21, X22
+	ADD	X12, X22, X12	// z_hi[1] = x[1] * y + z[1]
+	ADD	X21, X29, X13	// z_lo[1] = x[1] * y + z[1] + c
+	SLTU	X21, X13, X22
+	ADD	X12, X22, X29	// next c
+
+	MULHU	X14, X6, X15	// z_hi[2] = x[2] * y
+	MUL	X14, X6, X14	// z_lo[2] = x[2] * y
+	ADD	X14, X16, X21	// z_lo[2] = x[2] * y + z[2]
+	SLTU	X14, X21, X22
+	ADD	X15, X22, X15	// z_hi[2] = x[2] * y + z[2]
+	ADD	X21, X29, X16	// z_lo[2] = x[2] * y + z[2] + c
+	SLTU	X21, X16, X22
+	ADD	X15, X22, X29	// next c
+
+	MULHU	X17, X6, X18	// z_hi[3] = x[3] * y
+	MUL	X17, X6, X17	// z_lo[3] = x[3] * y
+	ADD	X17, X19, X21	// z_lo[3] = x[3] * y + z[3]
+	SLTU	X17, X21, X22
+	ADD	X18, X22, X18	// z_hi[3] = x[3] * y + z[3]
+	ADD	X21, X29, X19	// z_lo[3] = x[3] * y + z[3] + c
+	SLTU	X21, X19, X22
+	ADD	X18, X22, X29	// next c
+
+	MOV	X10, 0*8(X5)	// z[0]
+	MOV	X13, 1*8(X5)	// z[1]
+	MOV	X16, 2*8(X5)	// z[2]
+	MOV	X19, 3*8(X5)	// z[3]
+
+	ADDI	$32, X5
+	ADDI	$32, X7
+
+	ADDI	$-4, X30
+	BNEZ	X30, loop
+
+done:
+	MOV	X29, c+24(FP)
+	RET