s390x: decode #19

emmansun · Oct 17, 2024 · 2ef02a2 · 2ef02a2
1 parent 11e4128
commit 2ef02a2
Show file tree

Hide file tree

Showing 6 changed files with 151 additions and 51 deletions.
diff --git a/base64_asm.go b/base64_asm.go
@@ -2,7 +2,7 @@
 // Use of this source code is governed by a BSD 3-Clause-style
 // license that can be found in the LICENSE file.
 
-//go:build (amd64 || ppc64 || ppc64le) && !purego
+//go:build (amd64 || ppc64 || ppc64le || s390x) && !purego
 
 package base64
 

diff --git a/base64_asm_test.go b/base64_asm_test.go
@@ -2,7 +2,7 @@
 // Use of this source code is governed by a BSD 3-Clause-style
 // license that can be found in the LICENSE file.
 
-//go:build (amd64 || ppc64 || ppc64le) && !purego
+//go:build (amd64 || ppc64 || ppc64le || s390x) && !purego
 
 package base64
 

diff --git a/base64_generic.go b/base64_generic.go
@@ -1,4 +1,4 @@
-//go:build !(amd64 || arm64 || ppc64 || ppc64le) || purego
+//go:build !(amd64 || arm64 || ppc64 || ppc64le || s390x) || purego
 
 package base64
 

diff --git a/base64_s390x.go b/base64_s390x.go
diff --git a/base64_s390x.s b/base64_s390x.s
@@ -20,6 +20,26 @@ DATA base64_const<>+0x50(SB)/8, $0x0100001001000010 // mullo const
 DATA base64_const<>+0x58(SB)/8, $0x0100001001000010
 GLOBL base64_const<>(SB), (NOPTR+RODATA), $96
 
+DATA decode_const<>+0x00(SB)/8, $0x1010010204080408 // standard decode lut hi
+DATA decode_const<>+0x08(SB)/8, $0x1010101010101010
+DATA decode_const<>+0x10(SB)/8, $0x1511111111111111 // standard decode lut lo
+DATA decode_const<>+0x18(SB)/8, $0x1111131A1B1B1B1A
+DATA decode_const<>+0x20(SB)/8, $0x00101304BFBFB9B9 // standard decode lut roll
+DATA decode_const<>+0x28(SB)/8, $0x0000000000000000
+DATA decode_const<>+0x30(SB)/8, $0x4001400140014001 // reshuffle const0
+DATA decode_const<>+0x38(SB)/8, $0x4001400140014001
+DATA decode_const<>+0x40(SB)/8, $0x1000000110000001 // reshuffle const1
+DATA decode_const<>+0x48(SB)/8, $0x1000000110000001
+DATA decode_const<>+0x50(SB)/8, $0x010203050607090a // reshuffle mask
+DATA decode_const<>+0x58(SB)/8, $0x0b0d0e0f00000000
+DATA decode_const<>+0x60(SB)/8, $0x1010010204080428 // url decode lut hi
+DATA decode_const<>+0x68(SB)/8, $0x1010101010101010
+DATA decode_const<>+0x70(SB)/8, $0x1511111111111111 // url decode lut lo
+DATA decode_const<>+0x78(SB)/8, $0x1111131B1B1A1B33
+DATA decode_const<>+0x80(SB)/8, $0x00001104BFBFE0B9 // url decode lut roll
+DATA decode_const<>+0x88(SB)/8, $0xB900000000000000
+GLOBL decode_const<>(SB), (NOPTR+RODATA), $144
+
 #define REV_BYTES V0
 #define RESHUFFLE_MASK V1
 #define MULHI_MASK V2
@@ -58,7 +78,6 @@ TEXT ·encodeAsm(SB),NOSPLIT,$0
 	VREPIB $0xff, NEG
 	VZERO ZERO
 
-
 	MOVD $0, R4
 loop:
 	VL (R2), X0
@@ -99,6 +118,134 @@ done:
 #undef RANGE0_END
 #undef LUT
 #undef ZERO
+#undef NEG
 #undef X0
 #undef X1
 #undef X2
+
+#define LUT_HI V0
+#define LUT_LO V1
+#define LUT_ROLL V2
+#define RESHUFFLE_CONST0 V3
+#define RESHUFFLE_CONST1 V4
+#define RESHUFFLE_MASK V5
+#define ZERO V6
+#define NEG V7
+#define NIBBLE_MASK V8
+#define DECODE_CONST V9
+#define X0 V10
+#define X1 V11
+#define X2 V12
+#define X3 V13
+
+//func decodeStdAsm(dst, src []byte) int
+TEXT ·decodeStdAsm(SB),NOSPLIT,$0
+	MOVD dst_base+0(FP), R1
+	MOVD src_base+24(FP), R2
+	MOVD src_len+32(FP), R3
+
+	MOVD $decode_const<>(SB), R4
+	VLM (R4), LUT_HI, RESHUFFLE_MASK
+	VREPIB $0x0f, NIBBLE_MASK
+	VREPIB $0x2f, DECODE_CONST
+	VZERO ZERO
+
+loop:
+		VL (R2), X0
+		// validate the input
+		VESRLF $4, X0, X1
+		VN X1, NIBBLE_MASK, X1 // high nibbles
+		VN X0, NIBBLE_MASK, X2
+		VPERM LUT_HI, LUT_HI, X1, X3
+		VPERM LUT_LO, LUT_LO, X2, X2
+		VN X2, X3, X2
+		VCEQGS ZERO, X2, X2
+		BNE done
+
+		// decode the input
+		VCEQB DECODE_CONST, X0, X2
+		VAB X2, X1, X1
+		VPERM LUT_ROLL, LUT_ROLL, X1, X2
+		VAB X0, X2, X0
+
+		VMLEB RESHUFFLE_CONST0, X0, X1
+		VMLOB RESHUFFLE_CONST0, X0, X2
+		VAH X1, X2, X0
+		VMLEH RESHUFFLE_CONST1, X0, X1
+		VMLOH RESHUFFLE_CONST1, X0, X2
+		VAF X1, X2, X0
+
+		VPERM X0, X0, RESHUFFLE_MASK, X0
+		VST X0, (R1)
+
+		LAY 16(R2), R2
+		LAY 12(R1), R1
+		SUB $16, R3
+		CMPBGE R3, $24, loop
+
+done:
+	MOVD R3, ret+48(FP)
+	RET
+#undef LUT_HI
+#undef LUT_LO
+#undef LUT_ROLL
+#undef RESHUFFLE_CONST0
+#undef RESHUFFLE_CONST1
+#undef RESHUFFLE_MASK
+
+#define RESHUFFLE_CONST0 V0
+#define RESHUFFLE_CONST1 V1
+#define RESHUFFLE_MASK V2
+#define LUT_HI V3
+#define LUT_LO V4
+#define LUT_ROLL V5
+
+//func decodeUrlAsm(dst, src []byte) int
+TEXT ·decodeUrlAsm(SB),NOSPLIT,$0
+	MOVD dst_base+0(FP), R1
+	MOVD src_base+24(FP), R2
+	MOVD src_len+32(FP), R3
+
+	MOVD $decode_const<>(SB), R4
+	VLM (R4), LUT_HI, RESHUFFLE_MASK
+	VREPIB $0xff, NEG
+	VREPIB $0x0f, NIBBLE_MASK
+	VREPIB $0x5e, DECODE_CONST
+	VZERO ZERO
+
+loop:
+		VL (R2), X0
+		// validate the input
+		VESRLF $4, X0, X1
+		VN X1, NIBBLE_MASK, X1 // high nibbles
+		VN X0, NIBBLE_MASK, X2
+		VPERM LUT_HI, LUT_HI, X1, X3
+		VPERM LUT_LO, LUT_LO, X2, X2
+		VN X2, X3, X2
+		VCEQGS ZERO, X2, X2
+		BNE done
+
+		// decode the input
+		VCGTB(X0, DECODE_CONST, X2)
+		VSB X2, X1, X1
+		VPERM LUT_ROLL, LUT_ROLL, X1, X2
+		VAB X0, X2, X0
+
+		VMLEB RESHUFFLE_CONST0, X0, X1
+		VMLOB RESHUFFLE_CONST0, X0, X2
+		VAH X1, X2, X0
+		VMLEH RESHUFFLE_CONST1, X0, X1
+		VMLOH RESHUFFLE_CONST1, X0, X2
+		VAF X1, X2, X0
+
+		VPERM X0, X0, RESHUFFLE_MASK, X0
+		VST X0, (R1)
+
+		LAY 16(R2), R2
+		LAY 12(R1), R1
+		SUB $16, R3
+		CMPBGE R3, $24, loop
+
+done:
+	MOVD R3, ret+48(FP)
+	RET
diff --git a/base64_s390x_test.go b/base64_s390x_test.go