aboutsummaryrefslogtreecommitdiff
path: root/src/crypto
diff options
context:
space:
mode:
authorXiaolin Zhao <zhaoxiaolin@loongson.cn>2024-06-19 15:09:21 +0800
committerabner chenc <chenguoqi@loongson.cn>2024-09-04 00:31:27 +0000
commitdebfcb5ad87b276318bd6b725797e6808adeeae0 (patch)
tree2d9cc90180fab45a21b314e6dee7fdb5fc46777d /src/crypto
parent0c1627812406a76ac256a08bee985a9817372446 (diff)
downloadgo-debfcb5ad87b276318bd6b725797e6808adeeae0.tar.xz
crypto/internal/bigmod: provide assembly addMulVVW* for loong64
goos: linux goarch: loong64 pkg: crypto/internal/bigmod cpu: Loongson-3A6000 @ 2500.00MHz │ bench.old │ bench.new │ │ sec/op │ sec/op vs base │ ModAdd 159.5n ± 0% 159.8n ± 0% +0.19% (p=0.000 n=20) ModSub 161.5n ± 0% 161.7n ± 0% +0.12% (p=0.038 n=20) MontgomeryRepr 4.126µ ± 0% 2.932µ ± 0% -28.94% (p=0.000 n=20) MontgomeryMul 4.144µ ± 0% 2.930µ ± 0% -29.30% (p=0.000 n=20) ModMul 8.331µ ± 0% 5.956µ ± 0% -28.51% (p=0.000 n=20) ExpBig 11.65m ± 0% 11.64m ± 0% -0.04% (p=0.000 n=20) Exp 11.015m ± 0% 7.860m ± 0% -28.65% (p=0.000 n=20) geomean 17.34µ 14.28µ -17.64% goos: linux goarch: loong64 pkg: crypto/internal/bigmod cpu: Loongson-3A5000 @ 2500.00MHz │ bench.old │ bench.new │ │ sec/op │ sec/op vs base │ ModAdd 211.3n ± 0% 213.9n ± 0% +1.23% (p=0.000 n=20) ModSub 210.6n ± 0% 207.2n ± 0% -1.61% (p=0.000 n=20) MontgomeryRepr 5.442µ ± 0% 3.825µ ± 0% -29.71% (p=0.000 n=20) MontgomeryMul 5.379µ ± 0% 4.011µ ± 0% -25.43% (p=0.000 n=20) ModMul 10.868µ ± 0% 7.859µ ± 0% -27.69% (p=0.000 n=20) ExpBig 14.64m ± 0% 14.63m ± 0% -0.06% (p=0.035 n=20) Exp 14.39m ± 0% 10.38m ± 0% -27.86% (p=0.000 n=20) geomean 22.57µ 18.74µ -16.96% Change-Id: Id6ddc9552494e2a26e1a123f38e22d18bb78fdad Reviewed-on: https://go-review.googlesource.com/c/go/+/593595 Reviewed-by: Roland Shoemaker <roland@golang.org> Reviewed-by: Qiqi Huang <huangqiqi@loongson.cn> Reviewed-by: Filippo Valsorda <filippo@golang.org> Reviewed-by: abner chenc <chenguoqi@loongson.cn> LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Meidan Li <limeidan@loongson.cn> Reviewed-by: Dmitri Shuralyov <dmitshur@google.com>
Diffstat (limited to 'src/crypto')
-rw-r--r--src/crypto/internal/bigmod/nat_asm.go2
-rw-r--r--src/crypto/internal/bigmod/nat_loong64.s93
-rw-r--r--src/crypto/internal/bigmod/nat_noasm.go2
3 files changed, 95 insertions, 2 deletions
diff --git a/src/crypto/internal/bigmod/nat_asm.go b/src/crypto/internal/bigmod/nat_asm.go
index 0283b07e68..dd5419cb91 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 || riscv64 || s390x)
+//go:build !purego && (386 || amd64 || arm || arm64 || loong64 || ppc64 || ppc64le || riscv64 || s390x)
package bigmod
diff --git a/src/crypto/internal/bigmod/nat_loong64.s b/src/crypto/internal/bigmod/nat_loong64.s
new file mode 100644
index 0000000000..3423bd0b17
--- /dev/null
+++ b/src/crypto/internal/bigmod/nat_loong64.s
@@ -0,0 +1,93 @@
+// Copyright 2024 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.
+
+// derived from crypto/internal/bigmod/nat_riscv64.s
+
+//go:build !purego
+
+#include "textflag.h"
+
+// func addMulVVW1024(z, x *uint, y uint) (c uint)
+TEXT ·addMulVVW1024(SB),$0-32
+ MOVV $16, R8
+ JMP addMulVVWx(SB)
+
+// func addMulVVW1536(z, x *uint, y uint) (c uint)
+TEXT ·addMulVVW1536(SB),$0-32
+ MOVV $24, R8
+ JMP addMulVVWx(SB)
+
+// func addMulVVW2048(z, x *uint, y uint) (c uint)
+TEXT ·addMulVVW2048(SB),$0-32
+ MOVV $32, R8
+ JMP addMulVVWx(SB)
+
+TEXT addMulVVWx(SB),NOFRAME|NOSPLIT,$0
+ MOVV z+0(FP), R4
+ MOVV x+8(FP), R6
+ MOVV y+16(FP), R5
+ MOVV $0, R7
+
+ BEQ R8, R0, done
+loop:
+ MOVV 0*8(R4), R9 // z[0]
+ MOVV 1*8(R4), R10 // z[1]
+ MOVV 2*8(R4), R11 // z[2]
+ MOVV 3*8(R4), R12 // z[3]
+
+ MOVV 0*8(R6), R13 // x[0]
+ MOVV 1*8(R6), R14 // x[1]
+ MOVV 2*8(R6), R15 // x[2]
+ MOVV 3*8(R6), R16 // x[3]
+
+ MULHVU R13, R5, R17 // z_hi[0] = x[0] * y
+ MULV R13, R5, R13 // z_lo[0] = x[0] * y
+ ADDV R13, R9, R18 // z_lo[0] = x[0] * y + z[0]
+ SGTU R13, R18, R19
+ ADDV R17, R19, R17 // z_hi[0] = x[0] * y + z[0]
+ ADDV R18, R7, R9 // z_lo[0] = x[0] * y + z[0] + c
+ SGTU R18, R9, R19
+ ADDV R17, R19, R7 // next c
+
+ MULHVU R14, R5, R24 // z_hi[1] = x[1] * y
+ MULV R14, R5, R14 // z_lo[1] = x[1] * y
+ ADDV R14, R10, R18 // z_lo[1] = x[1] * y + z[1]
+ SGTU R14, R18, R19
+ ADDV R24, R19, R24 // z_hi[1] = x[1] * y + z[1]
+ ADDV R18, R7, R10 // z_lo[1] = x[1] * y + z[1] + c
+ SGTU R18, R10, R19
+ ADDV R24, R19, R7 // next c
+
+ MULHVU R15, R5, R25 // z_hi[2] = x[2] * y
+ MULV R15, R5, R15 // z_lo[2] = x[2] * y
+ ADDV R15, R11, R18 // z_lo[2] = x[2] * y + z[2]
+ SGTU R15, R18, R19
+ ADDV R25, R19, R25 // z_hi[2] = x[2] * y + z[2]
+ ADDV R18, R7, R11 // z_lo[2] = x[2] * y + z[2] + c
+ SGTU R18, R11, R19
+ ADDV R25, R19, R7 // next c
+
+ MULHVU R16, R5, R26 // z_hi[3] = x[3] * y
+ MULV R16, R5, R16 // z_lo[3] = x[3] * y
+ ADDV R16, R12, R18 // z_lo[3] = x[3] * y + z[3]
+ SGTU R16, R18, R19
+ ADDV R26, R19, R26 // z_hi[3] = x[3] * y + z[3]
+ ADDV R18, R7, R12 // z_lo[3] = x[3] * y + z[3] + c
+ SGTU R18, R12, R19
+ ADDV R26, R19, R7 // next c
+
+ MOVV R9, 0*8(R4) // z[0]
+ MOVV R10, 1*8(R4) // z[1]
+ MOVV R11, 2*8(R4) // z[2]
+ MOVV R12, 3*8(R4) // z[3]
+
+ ADDV $32, R4
+ ADDV $32, R6
+
+ SUBV $4, R8
+ BNE R8, R0, loop
+
+done:
+ MOVV R7, c+24(FP)
+ RET
diff --git a/src/crypto/internal/bigmod/nat_noasm.go b/src/crypto/internal/bigmod/nat_noasm.go
index 71f38da754..2501a6fb4c 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 || riscv64 || s390x)
+//go:build purego || !(386 || amd64 || arm || arm64 || loong64 || ppc64 || ppc64le || riscv64 || s390x)
package bigmod