math/big: minor tweaks to assembly code (slightly better performance)

Benchmarks run on 2.8GHz Quad-Code Intel Xeon,
4GB 800MHz DDR2 FB-DIMM ("PowerMac").

benchmark             old ns/op    new ns/op    delta
BenchmarkAddVV_1              7            7   -0.82%
BenchmarkAddVV_2              8            8   -3.46%
BenchmarkAddVV_3             10            9   -4.81%
BenchmarkAddVV_4              9            9   -1.89%
BenchmarkAddVV_5             11           10   -5.22%
BenchmarkAddVV_1e1           17           18   +4.05%
BenchmarkAddVV_1e2          117          115   -1.71%
BenchmarkAddVV_1e3         1095         1090   -0.46%
BenchmarkAddVV_1e4        13149        12679   -3.57%
BenchmarkAddVV_1e5       135133       129482   -4.18%
BenchmarkAddVW_1              6            6   -1.14%
BenchmarkAddVW_2              7            7   +3.78%
BenchmarkAddVW_3              8            8   +0.12%
BenchmarkAddVW_4              8            8   -6.52%
BenchmarkAddVW_5              9            8   -3.70%
BenchmarkAddVW_1e1           14           13   -4.29%
BenchmarkAddVW_1e2           97           96   -1.33%
BenchmarkAddVW_1e3          953          940   -1.36%
BenchmarkAddVW_1e4         9776         9527   -2.55%
BenchmarkAddVW_1e5       102396        97738   -4.55%

benchmark              old MB/s     new MB/s  speedup
BenchmarkAddVV_1        8702.84      8774.56    1.01x
BenchmarkAddVV_2       14739.60     15277.82    1.04x
BenchmarkAddVV_3       18375.37     19398.16    1.06x
BenchmarkAddVV_4       26935.44     27464.68    1.02x
BenchmarkAddVV_5       27754.04     29423.30    1.06x
BenchmarkAddVV_1e1     37050.89     35629.72    0.96x
BenchmarkAddVV_1e2     54289.15     55533.24    1.02x
BenchmarkAddVV_1e3     58428.83     58682.53    1.00x
BenchmarkAddVV_1e4     48670.55     50475.99    1.04x
BenchmarkAddVV_1e5     47360.54     49427.66    1.04x
BenchmarkAddVW_1       10397.27     10502.23    1.01x
BenchmarkAddVW_2       17279.03     16654.13    0.96x
BenchmarkAddVW_3       23858.39     23825.89    1.00x
BenchmarkAddVW_4       29799.42     31895.06    1.07x
BenchmarkAddVW_5       34781.83     36105.11    1.04x
BenchmarkAddVW_1e1     45629.88     47597.42    1.04x
BenchmarkAddVW_1e2     65341.93     66240.04    1.01x
BenchmarkAddVW_1e3     67153.67     68069.83    1.01x
BenchmarkAddVW_1e4     65464.60     67173.83    1.03x
BenchmarkAddVW_1e5     62501.88     65480.66    1.05x

R=iant
CC=golang-dev
https://golang.org/cl/6484056

src/pkg/math/big/arith_amd64.s

@@ -32,45 +32,44 @@ TEXT ·addVV(SB),7,$0
 	MOVQ z+0(FP), R10
 
 	MOVQ $0, CX		// c = 0
-	MOVQ $0, SI	     // i = 0
+	MOVQ $0, SI		// i = 0
 
 	// uncomment the next line to disable the unrolled loop
 	// JMP V1
 	
-	CMPQ DI, $4
-	JL V1			// if n < 4 goto V1
+	SUBQ $4, DI		// n -= 4
+	JL V1			// if n < 0 goto V1
 
-U1:	// n >= 4
+U1:	// n >= 0
 	// regular loop body unrolled 4x
+	RCRQ $1, CX		// CF = c
 	MOVQ 0(R8)(SI*8), R11
 	MOVQ 8(R8)(SI*8), R12
 	MOVQ 16(R8)(SI*8), R13
 	MOVQ 24(R8)(SI*8), R14
-	RCRQ $1, CX		// restore CF
 	ADCQ 0(R9)(SI*8), R11
 	ADCQ 8(R9)(SI*8), R12
 	ADCQ 16(R9)(SI*8), R13
 	ADCQ 24(R9)(SI*8), R14
-	RCLQ $1, CX		// save CF
 	MOVQ R11, 0(R10)(SI*8)
 	MOVQ R12, 8(R10)(SI*8)
 	MOVQ R13, 16(R10)(SI*8)
 	MOVQ R14, 24(R10)(SI*8)
+	RCLQ $1, CX		// c = CF
 
 	ADDQ $4, SI		// i += 4
 	SUBQ $4, DI		// n -= 4
-	CMPQ DI, $4
-	JGE U1			// if n >= 4 goto U1
+	JGE U1			// if n >= 0 goto U1
 
-V1:	CMPQ DI, $0
+V1:	ADDQ $4, DI		// n += 4
 	JLE E1			// if n <= 0 goto E1
 
 L1:	// n > 0
+	RCRQ $1, CX		// CF = c
 	MOVQ 0(R8)(SI*8), R11
-	RCRQ $1, CX		// restore CF
 	ADCQ 0(R9)(SI*8), R11
-	RCLQ $1, CX		// save CF
 	MOVQ R11, 0(R10)(SI*8)
+	RCLQ $1, CX		// c = CF
 
 	ADDQ $1, SI		// i++
 	SUBQ $1, DI		// n--
@@ -94,40 +93,39 @@ TEXT ·subVV(SB),7,$0
 	// uncomment the next line to disable the unrolled loop
 	// JMP V2
 	
-	CMPQ DI, $4
-	JL V2			// if n < 4 goto V2
+	SUBQ $4, DI		// n -= 4
+	JL V2			// if n < 0 goto V2
 
-U2:	// n >= 4
+U2:	// n >= 0
 	// regular loop body unrolled 4x
+	RCRQ $1, CX		// CF = c
 	MOVQ 0(R8)(SI*8), R11
 	MOVQ 8(R8)(SI*8), R12
 	MOVQ 16(R8)(SI*8), R13
 	MOVQ 24(R8)(SI*8), R14
-	RCRQ $1, CX		// restore CF
 	SBBQ 0(R9)(SI*8), R11
 	SBBQ 8(R9)(SI*8), R12
 	SBBQ 16(R9)(SI*8), R13
 	SBBQ 24(R9)(SI*8), R14
-	RCLQ $1, CX		// save CF
 	MOVQ R11, 0(R10)(SI*8)
 	MOVQ R12, 8(R10)(SI*8)
 	MOVQ R13, 16(R10)(SI*8)
 	MOVQ R14, 24(R10)(SI*8)
+	RCLQ $1, CX		// c = CF
 
 	ADDQ $4, SI		// i += 4
 	SUBQ $4, DI		// n -= 4
-	CMPQ DI, $4
-	JGE U2			// if n >= 4 goto U2
+	JGE U2			// if n >= 0 goto U2
 
-V2:	CMPQ DI, $0
+V2:	ADDQ $4, DI		// n += 4
 	JLE E2			// if n <= 0 goto E2
 
 L2:	// n > 0
+	RCRQ $1, CX		// CF = c
 	MOVQ 0(R8)(SI*8), R11
-	RCRQ $1, CX		// restore CF
 	SBBQ 0(R9)(SI*8), R11
-	RCLQ $1, CX		// save CF
 	MOVQ R11, 0(R10)(SI*8)
+	RCLQ $1, CX		// c = CF
 
 	ADDQ $1, SI		// i++
 	SUBQ $1, DI		// n--
@@ -149,10 +147,10 @@ TEXT ·addVW(SB),7,$0
 	// uncomment the next line to disable the unrolled loop
 	// JMP V3
 	
-	CMPQ DI, $4
+	SUBQ $4, DI		// n -= 4
 	JL V3			// if n < 4 goto V3
 
-U3:	// n >= 4
+U3:	// n >= 0
 	// regular loop body unrolled 4x
 	MOVQ 0(R8)(SI*8), R11
 	MOVQ 8(R8)(SI*8), R12
@@ -162,25 +160,24 @@ U3:	// n >= 4
 	ADCQ $0, R12
 	ADCQ $0, R13
 	ADCQ $0, R14
-	RCLQ $1, CX
-	ANDQ $1, CX
 	MOVQ R11, 0(R10)(SI*8)
 	MOVQ R12, 8(R10)(SI*8)
 	MOVQ R13, 16(R10)(SI*8)
 	MOVQ R14, 24(R10)(SI*8)
+	RCLQ $1, CX		// c = CF
+	ANDQ $1, CX
 
 	ADDQ $4, SI		// i += 4
 	SUBQ $4, DI		// n -= 4
-	CMPQ DI, $4
-	JGE U3			// if n >= 4 goto U3
+	JGE U3			// if n >= 0 goto U3
 
-V3:	CMPQ DI, $0
+V3:	ADDQ $4, DI		// n += 4
 	JLE E3			// if n <= 0 goto E3
 
 L3:	// n > 0
 	ADDQ 0(R8)(SI*8), CX
 	MOVQ CX, 0(R10)(SI*8)
-	RCLQ $1, CX
+	RCLQ $1, CX		// c = CF
 	ANDQ $1, CX
 
 	ADDQ $1, SI		// i++
@@ -204,10 +201,10 @@ TEXT ·subVW(SB),7,$0
 	// uncomment the next line to disable the unrolled loop
 	// JMP V4
 	
-	CMPQ DI, $4
+	SUBQ $4, DI		// n -= 4
 	JL V4			// if n < 4 goto V4
 
-U4:	// n >= 4
+U4:	// n >= 0
 	// regular loop body unrolled 4x
 	MOVQ 0(R8)(SI*8), R11
 	MOVQ 8(R8)(SI*8), R12
@@ -217,26 +214,25 @@ U4:	// n >= 4
 	SBBQ $0, R12
 	SBBQ $0, R13
 	SBBQ $0, R14
-	RCLQ $1, CX
-	ANDQ $1, CX
 	MOVQ R11, 0(R10)(SI*8)
 	MOVQ R12, 8(R10)(SI*8)
 	MOVQ R13, 16(R10)(SI*8)
 	MOVQ R14, 24(R10)(SI*8)
+	RCLQ $1, CX		// c = CF
+	ANDQ $1, CX
 
 	ADDQ $4, SI		// i += 4
 	SUBQ $4, DI		// n -= 4
-	CMPQ DI, $4
-	JGE U4			// if n >= 4 goto U4
+	JGE U4			// if n >= 0 goto U4
 
-V4:	CMPQ DI, $0
+V4:	ADDQ $4, DI		// n += 4
 	JLE E4			// if n <= 0 goto E4
 
 L4:	// n > 0
 	MOVQ 0(R8)(SI*8), R11
 	SUBQ CX, R11
 	MOVQ R11, 0(R10)(SI*8)
-	RCLQ $1, CX
+	RCLQ $1, CX		// c = CF
 	ANDQ $1, CX
 
 	ADDQ $1, SI		// i++