s2: Improve Better compression slightly (#663)

* Do forward hashing of second index.
klauspost · May 4, 2023 · 9a951be · 9a951be
1 parent aa3702a
commit 9a951be
Show file tree

Hide file tree

Showing 4 changed files with 146 additions and 111 deletions.
diff --git a/s2/_generate/gen.go b/s2/_generate/gen.go
@@ -1417,24 +1417,30 @@ func (o options) genEncodeBetterBlockAsm(name string, lTableBits, sTableBits, sk
 		MOVL(plusone0.As32(), sTab.Idx(hash0s, 4))
 		MOVL(plusone1.As32(), sTab.Idx(hash1s, 4))
 
+		// index2 := (index0 + index1 + 1) >> 1
+		index2 := GP64()
+		LEAQ(Mem{Base: index1, Disp: 1, Index: index0, Scale: 1}, index2)
+		SHRQ(U8(1), index2)
+
 		ADDQ(U8(1), index0)
 		SUBQ(U8(1), index1)
 
 		Label("index_loop_" + name)
-		CMPQ(index0, index1)
+		// for index2 < index1
+		CMPQ(index2, index1)
 		JAE(LabelRef("search_loop_" + name))
 		hash0l, hash1l = GP64(), GP64()
 		MOVQ(Mem{Base: src, Index: index0, Scale: 1, Disp: 0}, hash0l)
-		MOVQ(Mem{Base: src, Index: index1, Scale: 1, Disp: 0}, hash1l)
+		MOVQ(Mem{Base: src, Index: index2, Scale: 1, Disp: 0}, hash1l)
 
 		lHasher.hash(hash0l)
 		lHasher.hash(hash1l)
 
 		MOVL(index0.As32(), lTab.Idx(hash0l, 4))
-		MOVL(index1.As32(), lTab.Idx(hash1l, 4))
+		MOVL(index2.As32(), lTab.Idx(hash1l, 4))
 
 		ADDQ(U8(2), index0)
-		SUBQ(U8(2), index1)
+		ADDQ(U8(2), index2)
 		JMP(LabelRef("index_loop_" + name))
 	} else {
 		lHasher := hashN(lHashBytes, lTableBits)

diff --git a/s2/encode_better.go b/s2/encode_better.go
@@ -268,18 +268,21 @@ func encodeBlockBetterGo(dst, src []byte) (d int) {
 		lTable[hash7(cv0, lTableBits)] = uint32(index0)
 		sTable[hash4(cv0>>8, sTableBits)] = uint32(index0 + 1)
 
+		// lTable could be postponed, but very minor difference.
 		lTable[hash7(cv1, lTableBits)] = uint32(index1)
 		sTable[hash4(cv1>>8, sTableBits)] = uint32(index1 + 1)
 		index0 += 1
 		index1 -= 1
 		cv = load64(src, s)
 
-		// index every second long in between.
-		for index0 < index1 {
+		// Index large values sparsely in between.
+		// We do two starting from different offsets for speed.
+		index2 := (index0 + index1 + 1) >> 1
+		for index2 < index1 {
 			lTable[hash7(load64(src, index0), lTableBits)] = uint32(index0)
-			lTable[hash7(load64(src, index1), lTableBits)] = uint32(index1)
+			lTable[hash7(load64(src, index2), lTableBits)] = uint32(index2)
 			index0 += 2
-			index1 -= 2
+			index2 += 2
 		}
 	}
 
@@ -458,12 +461,14 @@ func encodeBlockBetterSnappyGo(dst, src []byte) (d int) {
 		index1 -= 1
 		cv = load64(src, s)
 
-		// index every second long in between.
-		for index0 < index1 {
+		// Index large values sparsely in between.
+		// We do two starting from different offsets for speed.
+		index2 := (index0 + index1 + 1) >> 1
+		for index2 < index1 {
 			lTable[hash7(load64(src, index0), lTableBits)] = uint32(index0)
-			lTable[hash7(load64(src, index1), lTableBits)] = uint32(index1)
+			lTable[hash7(load64(src, index2), lTableBits)] = uint32(index2)
 			index0 += 2
-			index1 -= 2
+			index2 += 2
 		}
 	}
 
@@ -863,12 +868,14 @@ searchDict:
 		index1 -= 1
 		cv = load64(src, s)
 
-		// index every second long in between.
-		for index0 < index1 {
+		// Index large values sparsely in between.
+		// We do two starting from different offsets for speed.
+		index2 := (index0 + index1 + 1) >> 1
+		for index2 < index1 {
 			lTable[hash7(load64(src, index0), lTableBits)] = uint32(index0)
-			lTable[hash7(load64(src, index1), lTableBits)] = uint32(index1)
+			lTable[hash7(load64(src, index2), lTableBits)] = uint32(index2)
 			index0 += 2
-			index1 -= 2
+			index2 += 2
 		}
 	}
 
@@ -1076,12 +1083,14 @@ searchDict:
 		index1 -= 1
 		cv = load64(src, s)
 
-		// index every second long in between.
-		for index0 < index1 {
+		// Index large values sparsely in between.
+		// We do two starting from different offsets for speed.
+		index2 := (index0 + index1 + 1) >> 1
+		for index2 < index1 {
 			lTable[hash7(load64(src, index0), lTableBits)] = uint32(index0)
-			lTable[hash7(load64(src, index1), lTableBits)] = uint32(index1)
+			lTable[hash7(load64(src, index2), lTableBits)] = uint32(index2)
 			index0 += 2
-			index1 -= 2
+			index2 += 2
 		}
 	}