Various updates for reading/writing streams; 'faster' benchmarks; doc…

… comments; gofmt

README.md

@@ -23,77 +23,75 @@ goarch: amd64
 pkg: github.com/theMPatel/streamvbyte-simdgo/pkg
 cpu: Intel(R) Core(TM) i7-8700B CPU @ 3.20GHz
 --
-BenchmarkMemCopy8Uint32-12    	448033621	         2.613 ns/op	12247.79 MB/s
+BenchmarkMemCopy8Uint32-12    	443656366	         2.621 ns/op	12209.73 MB/s
 
 goos: darwin
 goarch: amd64
 pkg: github.com/theMPatel/streamvbyte-simdgo/pkg/decode
 cpu: Intel(R) Core(TM) i7-8700B CPU @ 3.20GHz
 --
-BenchmarkGet8uint32Fast-12           	368871128	         3.234 ns/op	9895.19 MB/s
-BenchmarkGet8uint32DeltaFast-12      	306617568	         3.853 ns/op	8304.19 MB/s
-BenchmarkGet8uint32Scalar-12         	67530424	        18.59 ns/op	1721.51 MB/s
-BenchmarkGet8uint32DeltaScalar-12    	65821027	        19.17 ns/op	1668.90 MB/s
-BenchmarkGet8uint32Varint-12         	26296116	        46.45 ns/op	 688.88 MB/s
-BenchmarkGet8uint32DeltaVarint-12    	21070124	        57.95 ns/op	 552.16 MB/s
+BenchmarkGet8uint32Fast-12           	376997336	         3.254 ns/op	9832.71 MB/s
+BenchmarkGet8uint32DeltaFast-12      	310897352	         3.915 ns/op	8173.85 MB/s
+BenchmarkGet8uint32Scalar-12         	63194713	        18.70 ns/op	1710.88 MB/s
+BenchmarkGet8uint32DeltaScalar-12    	59567296	        20.09 ns/op	1592.49 MB/s
+BenchmarkGet8uint32Varint-12         	24679526	        51.44 ns/op	 622.12 MB/s
+BenchmarkGet8uint32DeltaVarint-12    	21775758	        54.76 ns/op	 584.42 MB/s
 
 goos: darwin
 goarch: amd64
 pkg: github.com/theMPatel/streamvbyte-simdgo/pkg/encode
 cpu: Intel(R) Core(TM) i7-8700B CPU @ 3.20GHz
 --
-BenchmarkPut8uint32Fast-12           	310744244	         3.883 ns/op	8241.67 MB/s
-BenchmarkPut8uint32DeltaFast-12      	276355396	         4.305 ns/op	7434.07 MB/s
-BenchmarkPut8uint32Scalar-12         	41064336	        29.89 ns/op	1070.47 MB/s
-BenchmarkPut8uint32DeltaScalar-12    	40731841	        29.17 ns/op	1096.87 MB/s
-BenchmarkPut8uint32Varint-12         	48454465	        22.42 ns/op	1427.45 MB/s
-BenchmarkPut8uint32DeltaVarint-12    	68148270	        22.23 ns/op	1439.20 MB/s
+BenchmarkPut8uint32Fast-12           	305809893	         3.943 ns/op	8115.94 MB/s
+BenchmarkPut8uint32DeltaFast-12      	275407464	         4.376 ns/op	7312.83 MB/s
+BenchmarkPut8uint32Scalar-12         	40038918	        28.51 ns/op	1122.44 MB/s
+BenchmarkPut8uint32DeltaScalar-12    	38668948	        30.92 ns/op	1035.07 MB/s
+BenchmarkPut8uint32Varint-12         	64186053	        20.79 ns/op	1539.23 MB/s
+BenchmarkPut8uint32DeltaVarint-12    	64914912	        20.13 ns/op	1589.52 MB/s
 
 goos: darwin
 goarch: amd64
 pkg: github.com/theMPatel/streamvbyte-simdgo/pkg/stream/reader
 cpu: Intel(R) Core(TM) i7-8700B CPU @ 3.20GHz
 --
-BenchmarkReadAllFast/Count_1e0-12         	100000000	        11.64 ns/op	 343.50 MB/s
-BenchmarkReadAllFast/Count_1e1-12         	27834266	        43.15 ns/op	 927.09 MB/s
-BenchmarkReadAllFast/Count_1e2-12         	11309079	       106.0 ns/op	3773.54 MB/s
-BenchmarkReadAllFast/Count_1e3-12         	 1673222	       720.6 ns/op	5550.61 MB/s
-BenchmarkReadAllFast/Count_1e4-12         	  171164	      6941 ns/op	5763.09 MB/s
-BenchmarkReadAllFast/Count_1e5-12         	   17080	     70616 ns/op	5664.44 MB/s
-BenchmarkReadAllFast/Count_1e6-12         	    1640	    709504 ns/op	5637.74 MB/s
-BenchmarkReadAllFast/Count_1e7-12         	     154	   7670283 ns/op	5214.93 MB/s
-BenchmarkFastRead-12                      	  421057	      2885 ns/op	5678.46 MB/s
-BenchmarkReadAllScalar/Count_1e0-12       	128565745	         9.355 ns/op	 427.59 MB/s
-BenchmarkReadAllScalar/Count_1e1-12       	37689912	        32.12 ns/op	1245.18 MB/s
-BenchmarkReadAllScalar/Count_1e2-12       	 4740482	       252.9 ns/op	1581.38 MB/s
-BenchmarkReadAllScalar/Count_1e3-12       	  482290	      2492 ns/op	1605.11 MB/s
-BenchmarkReadAllScalar/Count_1e4-12       	   17554	     68111 ns/op	 587.28 MB/s
-BenchmarkReadAllScalar/Count_1e5-12       	    1534	    776822 ns/op	 514.92 MB/s
-BenchmarkReadAllScalar/Count_1e6-12       	     153	   7792223 ns/op	 513.33 MB/s
-BenchmarkReadAllScalar/Count_1e7-12       	      14	  78339118 ns/op	 510.60 MB/s
+BenchmarkReadAllFast/Count_1e0-12         	100000000	        11.88 ns/op	 336.65 MB/s
+BenchmarkReadAllFast/Count_1e1-12         	23072438	        45.64 ns/op	 876.40 MB/s
+BenchmarkReadAllFast/Count_1e2-12         	11020350	       110.9 ns/op	3606.13 MB/s
+BenchmarkReadAllFast/Count_1e3-12         	 1636318	       726.5 ns/op	5506.11 MB/s
+BenchmarkReadAllFast/Count_1e4-12         	  172294	      7049 ns/op	5674.80 MB/s
+BenchmarkReadAllFast/Count_1e5-12         	   16779	     71174 ns/op	5620.04 MB/s
+BenchmarkReadAllFast/Count_1e6-12         	    1612	    723231 ns/op	5530.73 MB/s
+BenchmarkReadAllFast/Count_1e7-12         	     153	   7797910 ns/op	5129.58 MB/s
+BenchmarkReadAllScalar/Count_1e0-12       	129320632	         9.180 ns/op	 435.72 MB/s
+BenchmarkReadAllScalar/Count_1e1-12       	37140794	        32.83 ns/op	1218.47 MB/s
+BenchmarkReadAllScalar/Count_1e2-12       	 5020092	       240.7 ns/op	1661.61 MB/s
+BenchmarkReadAllScalar/Count_1e3-12       	  501602	      2408 ns/op	1661.22 MB/s
+BenchmarkReadAllScalar/Count_1e4-12       	   50671	     23728 ns/op	1685.80 MB/s
+BenchmarkReadAllScalar/Count_1e5-12       	    5133	    239372 ns/op	1671.04 MB/s
+BenchmarkReadAllScalar/Count_1e6-12       	     500	   2390896 ns/op	1673.01 MB/s
+BenchmarkReadAllScalar/Count_1e7-12       	      48	  24360062 ns/op	1642.03 MB/s
 
 goos: darwin
 goarch: amd64
 pkg: github.com/theMPatel/streamvbyte-simdgo/pkg/stream/writer
 cpu: Intel(R) Core(TM) i7-8700B CPU @ 3.20GHz
 --
-BenchmarkWriteAllFast/Count_1e0-12         	54667987	        21.98 ns/op	 181.97 MB/s
-BenchmarkWriteAllFast/Count_1e1-12         	28875541	        41.76 ns/op	 957.95 MB/s
-BenchmarkWriteAllFast/Count_1e2-12         	 7327003	       161.7 ns/op	2473.45 MB/s
-BenchmarkWriteAllFast/Count_1e3-12         	  992776	      1229 ns/op	3255.02 MB/s
-BenchmarkWriteAllFast/Count_1e4-12         	   98493	     11569 ns/op	3457.37 MB/s
-BenchmarkWriteAllFast/Count_1e5-12         	   10000	    108539 ns/op	3685.30 MB/s
-BenchmarkWriteAllFast/Count_1e6-12         	     915	   1197473 ns/op	3340.37 MB/s
-BenchmarkWriteAllFast/Count_1e7-12         	     106	  10155791 ns/op	3938.64 MB/s
-BenchmarkFastWrite-12                      	  247346	      4677 ns/op	3503.21 MB/s
-BenchmarkWriteAllScalar/Count_1e0-12       	57365708	        21.07 ns/op	 189.83 MB/s
-BenchmarkWriteAllScalar/Count_1e1-12       	18698986	        63.59 ns/op	 629.05 MB/s
-BenchmarkWriteAllScalar/Count_1e2-12       	 2709844	       440.5 ns/op	 908.03 MB/s
-BenchmarkWriteAllScalar/Count_1e3-12       	  286054	      4162 ns/op	 960.98 MB/s
-BenchmarkWriteAllScalar/Count_1e4-12       	   12038	     99717 ns/op	 401.14 MB/s
-BenchmarkWriteAllScalar/Count_1e5-12       	    1141	   1049287 ns/op	 381.21 MB/s
-BenchmarkWriteAllScalar/Count_1e6-12       	     100	  10509348 ns/op	 380.61 MB/s
-BenchmarkWriteAllScalar/Count_1e7-12       	      10	 105091277 ns/op	 380.62 MB/s
+BenchmarkWriteAllFast/Count_1e0-12         	54797875	        21.93 ns/op	 182.44 MB/s
+BenchmarkWriteAllFast/Count_1e1-12         	28054924	        41.96 ns/op	 953.26 MB/s
+BenchmarkWriteAllFast/Count_1e2-12         	 7326836	       161.7 ns/op	2473.55 MB/s
+BenchmarkWriteAllFast/Count_1e3-12         	  972864	      1239 ns/op	3227.39 MB/s
+BenchmarkWriteAllFast/Count_1e4-12         	   98656	     11702 ns/op	3418.09 MB/s
+BenchmarkWriteAllFast/Count_1e5-12         	   10000	    109297 ns/op	3659.74 MB/s
+BenchmarkWriteAllFast/Count_1e6-12         	     933	   1188199 ns/op	3366.44 MB/s
+BenchmarkWriteAllFast/Count_1e7-12         	     102	  10163143 ns/op	3935.79 MB/s
+BenchmarkWriteAllScalar/Count_1e0-12       	53847132	        21.88 ns/op	 182.84 MB/s
+BenchmarkWriteAllScalar/Count_1e1-12       	18313590	        65.17 ns/op	 613.79 MB/s
+BenchmarkWriteAllScalar/Count_1e2-12       	 2860240	       419.3 ns/op	 953.93 MB/s
+BenchmarkWriteAllScalar/Count_1e3-12       	  292951	      4099 ns/op	 975.82 MB/s
+BenchmarkWriteAllScalar/Count_1e4-12       	   29824	     40340 ns/op	 991.57 MB/s
+BenchmarkWriteAllScalar/Count_1e5-12       	    3045	    395439 ns/op	1011.53 MB/s
+BenchmarkWriteAllScalar/Count_1e6-12       	     297	   3945190 ns/op	1013.89 MB/s
+BenchmarkWriteAllScalar/Count_1e7-12       	      31	  41197829 ns/op	 970.92 MB/s
 ```
 
 A note on the benchmarks: An array of random uint32's is generated and then encoded/decoded over

pkg/stream/reader/reader.go

@@ -5,25 +5,77 @@ import (
 	"github.com/theMPatel/streamvbyte-simdgo/pkg/shared"
 )
 
+const (
+	jump     = 16
+	jumpCtrl = jump / 4
+)
+
+// ReadAll will read the entire input stream into out according to the
+// Stream VByte format. It will select the best implementation depending
+// on the presence of special hardware instructions.
+//
+// Note: It is your responsibility to ensure that the incoming slices are
+// appropriately sized as well as tracking the count of integers in the
+// stream.
+func ReadAll(count int, stream []byte, out []uint32) {
+	if decode.GetMode() == shared.Fast {
+		ReadAllFast(count, stream, out)
+	} else {
+		ReadAllScalar(count, stream, out)
+	}
+}
+
+// ReadAllScalar will read the entire input stream into out according to the
+// Stream VByte format.
+//
+// Note: It is your responsibility to ensure that the incoming slices are
+// appropriately sized as well as tracking the count of integers in the
+// stream.
 func ReadAllScalar(count int, stream []byte, out []uint32) {
 	var (
 		ctrlLen = (count + 3) / 4
 
-		dataPos = ctrlLen
-		ctrlPos = 0
-		numsPos = 0
-		lowest4 = count &^ 3
+		dataPos    = ctrlLen
+		ctrlPos    = 0
+		decoded    = 0
+		lowestJump = count &^ (jump - 1)
+		lowest4    = count &^ 3
 	)
 
-	for ; numsPos < lowest4; numsPos += 4 {
+	for ; decoded < lowestJump; decoded += jump {
+		data := stream[dataPos:]
+		ctrls := stream[ctrlPos : ctrlPos+jumpCtrl]
+		nums := out[decoded : decoded+jump]
+
+		ctrl := ctrls[0]
+		decode.Get4uint32Scalar(data, nums, ctrl)
+		sizeA := shared.ControlByteToSize(ctrl)
+
+		ctrl = ctrls[1]
+		decode.Get4uint32Scalar(data[sizeA:], nums[4:], ctrl)
+		sizeB := shared.ControlByteToSize(ctrl)
+
+		ctrl = ctrls[2]
+		decode.Get4uint32Scalar(data[sizeA+sizeB:], nums[8:], ctrl)
+		sizeC := shared.ControlByteToSize(ctrl)
+
+		ctrl = ctrls[3]
+		decode.Get4uint32Scalar(data[sizeA+sizeB+sizeC:], nums[12:], ctrl)
+		sizeD := shared.ControlByteToSize(ctrl)
+
+		dataPos += sizeA + sizeB + sizeC + sizeD
+		ctrlPos += jumpCtrl
+	}
+
+	for ; decoded < lowest4; decoded += 4 {
 		ctrl := stream[ctrlPos]
-		decode.Get4uint32Scalar(stream[dataPos:], out[numsPos:], ctrl)
+		decode.Get4uint32Scalar(stream[dataPos:], out[decoded:], ctrl)
 		size := shared.ControlByteToSize(ctrl)
 		dataPos += size
 		ctrlPos++
 	}
 
 	if lowest4 != count {
-		decode.GetUint32Scalar(stream[dataPos:], out[numsPos:], stream[ctrlPos], count-lowest4)
+		decode.GetUint32Scalar(stream[dataPos:], out[decoded:], stream[ctrlPos], count-lowest4)
 	}
 }

pkg/stream/reader/reader_amd64.go

@@ -7,6 +7,12 @@ import (
 	"github.com/theMPatel/streamvbyte-simdgo/pkg/shared"
 )
 
+// ReadAllFast will read the entire input stream into out according to the
+// Stream VByte format using special hardware instructions.
+//
+// Note: It is your responsibility to ensure that the incoming slices are
+// appropriately sized as well as tracking the count of integers in the
+// stream.
 func ReadAllFast(count int, stream []byte, out []uint32) {
 	var (
 		ctrlPos = 0

pkg/stream/reader/reader_base.go

@@ -0,0 +1,7 @@
+// +build !amd64
+
+package reader
+
+func ReadAllFast(count int, stream []byte, out []uint32) {
+	panic("unreachable")
+}

pkg/stream/reader/reader_test.go

@@ -68,26 +68,11 @@ func BenchmarkReadAllFast(b *testing.B) {
 
 var readSinkB []uint32
 
-func BenchmarkFastRead(b *testing.B) {
-	count := 4096
-	nums := util.GenUint32(count)
-	stream := writer.WriteAllScalar(nums)
-	per := count * encode.MaxBytesPerNum
-	out := make([]uint32, count)
-	b.SetBytes(int64(per))
-	b.ResetTimer()
-	for i := 0; i < b.N; i++ {
-		ReadAllFast(count, stream, out)
-	}
-	readSinkB = out
-}
-
-var readSinkC []uint32
-
 func BenchmarkReadAllScalar(b *testing.B) {
 	for i := 0; i < 8; i++ {
 		count := int(math.Pow10(i))
 		nums := util.GenUint32(count)
+		util.SortUint32(nums)
 		stream := writer.WriteAllScalar(nums)
 		out := make([]uint32, count)
 		b.Run(fmt.Sprintf("Count_1e%d", i), func(b *testing.B) {
@@ -96,39 +81,7 @@ func BenchmarkReadAllScalar(b *testing.B) {
 			for i := 0; i < b.N; i++ {
 				ReadAllScalar(count, stream, out)
 			}
-			readSinkC = out
+			readSinkB = out
 		})
 	}
 }
-
-var readSinkD []uint32
-
-func BenchmarkReadAllScalar1e4(b *testing.B) {
-	count := int(math.Pow10(4))
-	nums := util.GenUint32(count)
-	stream := writer.WriteAllScalar(nums)
-	out := make([]uint32, count)
-
-	// 68982
-	b.SetBytes(int64(count * encode.MaxBytesPerNum))
-	b.ResetTimer()
-	for i := 0; i < b.N; i++ {
-		ReadAllScalar(count, stream, out)
-	}
-	readSinkD = out
-}
-
-func BenchmarkReadAllScalar1e3(b *testing.B) {
-	count := int(math.Pow10(3))
-	nums := util.GenUint32(count)
-	stream := writer.WriteAllScalar(nums)
-	out := make([]uint32, count)
-
-	// 2555
-	b.SetBytes(int64(count * encode.MaxBytesPerNum))
-	b.ResetTimer()
-	for i := 0; i < b.N; i++ {
-		ReadAllScalar(count, stream, out)
-	}
-	readSinkC = out
-}

pkg/stream/writer/writer.go

@@ -5,20 +5,65 @@ import (
 	"github.com/theMPatel/streamvbyte-simdgo/pkg/shared"
 )
 
+const (
+	jump     = 16
+	jumpCtrl = jump / 4
+)
+
+// WriteAll will encode all the integers from in using the Stream VByte
+// format and will return the byte array holding the encoded data. It will
+// select the best implementation depending on the presence of special
+// hardware instructions.
+func WriteAll(in []uint32) []byte {
+	if encode.GetMode() == shared.Fast {
+		return WriteAllFast(in)
+	} else {
+		return WriteAllScalar(in)
+	}
+}
+
+// WriteAllScalar will encode all the integers from in using the Stream VByte
+// format and will return the byte array holding the encoded data.
 func WriteAllScalar(in []uint32) []byte {
 	var (
 		count   = len(in)
 		ctrlLen = (count + 3) / 4
 		stream  = make([]byte, ctrlLen+(encode.MaxBytesPerNum*count))
 
-		dataPos = ctrlLen
-		ctrlPos = 0
-		numsPos = 0
-		lowest4 = count &^ 3
+		dataPos    = ctrlLen
+		ctrlPos    = 0
+		encoded    = 0
+		lowestJump = count &^ (jump - 1)
+		lowest4    = count &^ 3
 	)
 
-	for ; numsPos < lowest4; numsPos += 4 {
-		ctrl := encode.Put4uint32Scalar(in[numsPos:], stream[dataPos:])
+	for ; encoded < lowestJump; encoded += jump {
+		nums := in[encoded : encoded+jump]
+		data := stream[dataPos:]
+		ctrls := stream[ctrlPos : ctrlPos+jumpCtrl]
+
+		ctrl := encode.Put4uint32Scalar(nums, data)
+		ctrls[0] = ctrl
+		sizeA := shared.ControlByteToSize(ctrl)
+
+		ctrl = encode.Put4uint32Scalar(nums[4:], data[sizeA:])
+		ctrls[1] = ctrl
+		sizeB := shared.ControlByteToSize(ctrl)
+
+		ctrl = encode.Put4uint32Scalar(nums[8:], data[sizeA+sizeB:])
+		ctrls[2] = ctrl
+		sizeC := shared.ControlByteToSize(ctrl)
+
+		ctrl = encode.Put4uint32Scalar(nums[12:], data[sizeA+sizeB+sizeC:])
+		ctrls[3] = ctrl
+		sizeD := shared.ControlByteToSize(ctrl)
+
+		dataPos += sizeA + sizeB + sizeC + sizeD
+		ctrlPos += jumpCtrl
+	}
+
+	for ; encoded < lowest4; encoded += 4 {
+		ctrl := encode.Put4uint32Scalar(in[encoded:], stream[dataPos:])
 		stream[ctrlPos] = ctrl
 		size := shared.ControlByteToSize(ctrl)
 		dataPos += size
@@ -27,7 +72,7 @@ func WriteAllScalar(in []uint32) []byte {
 
 	if lowest4 != count {
 		nums := count - lowest4
-		ctrl := encode.PutUint32Scalar(in[numsPos:], stream[dataPos:], nums)
+		ctrl := encode.PutUint32Scalar(in[encoded:], stream[dataPos:], nums)
 		size := shared.ControlByteToSize(ctrl)
 		size -= 4 - nums
 		dataPos += size

pkg/stream/writer/writer_amd64.go

@@ -7,6 +7,9 @@ import (
 	"github.com/theMPatel/streamvbyte-simdgo/pkg/shared"
 )
 
+// WriteAllFast will encode all the integers from in using the Stream VByte
+// format using special hardware instructions and will return the byte array
+// holding the encoded data.
 func WriteAllFast(in []uint32) []byte {
 	var (
 		count   = len(in)

pkg/stream/writer/writer_base.go

@@ -0,0 +1,7 @@
+// +build !amd64
+
+package writer
+
+func WriteAllFast(in []uint32) []byte {
+	panic("unreachable")
+}