diff --git a/s2/_generate/gen.go b/s2/_generate/gen.go
index c939174b05..5c9b5dc57e 100644
--- a/s2/_generate/gen.go
+++ b/s2/_generate/gen.go
@@ -237,7 +237,6 @@ func (o options) genEncodeBlockAsm(name string, tableBits, skipLog, hashBytes, m
 		})
 
 		cv := GP64()
-		MOVQ(Mem{Base: src, Index: s, Scale: 1}, cv)
 		nextS := GP32()
 		// nextS := s + (s-nextEmit)>>6 + 4
 		{
@@ -252,6 +251,7 @@ func (o options) genEncodeBlockAsm(name string, tableBits, skipLog, hashBytes, m
 			CMPL(nextS.As32(), sLimitL)
 			JGE(LabelRef("emit_remainder_" + name))
 		}
+		MOVQ(Mem{Base: src, Index: s, Scale: 1}, cv)
 		assert(func(ok LabelRef) {
 			// Check if s is valid (we should have jumped above if not)
 			tmp := GP64()
@@ -888,7 +888,6 @@ func (o options) genEncodeBetterBlockAsm(name string, lTableBits, skipLog, lHash
 		})
 
 		cv := GP64()
-		MOVQ(Mem{Base: src, Index: s, Scale: 1}, cv)
 		nextS := GP32()
 		// nextS := s + (s-nextEmit)>>skipLog + 1
 		{
@@ -903,6 +902,7 @@ func (o options) genEncodeBetterBlockAsm(name string, lTableBits, skipLog, lHash
 			CMPL(nextS.As32(), sLimitL)
 			JGE(LabelRef("emit_remainder_" + name))
 		}
+		MOVQ(Mem{Base: src, Index: s, Scale: 1}, cv)
 		assert(func(ok LabelRef) {
 			// Check if s is valid (we should have jumped above if not)
 			tmp := GP64()
diff --git a/s2/encodeblock_amd64.s b/s2/encodeblock_amd64.s
index 916a9736fc..7d1ed1bd9c 100644
--- a/s2/encodeblock_amd64.s
+++ b/s2/encodeblock_amd64.s
@@ -40,13 +40,13 @@ zero_loop_encodeBlockAsm:
 	MOVQ  src_base+24(FP), DX
 
 search_loop_encodeBlockAsm:
-	MOVQ  (DX)(CX*1), SI
 	MOVL  CX, BP
 	SUBL  12(SP), BP
 	SHRL  $0x06, BP
 	LEAL  4(CX)(BP*1), BP
 	CMPL  BP, 8(SP)
 	JGE   emit_remainder_encodeBlockAsm
+	MOVQ  (DX)(CX*1), SI
 	MOVL  BP, 20(SP)
 	MOVQ  $0x0000cf1bbcdcbf9b, R8
 	MOVQ  SI, R9
@@ -1252,13 +1252,13 @@ zero_loop_encodeBlockAsm4MB:
 	MOVQ  src_base+24(FP), DX
 
 search_loop_encodeBlockAsm4MB:
-	MOVQ  (DX)(CX*1), SI
 	MOVL  CX, BP
 	SUBL  12(SP), BP
 	SHRL  $0x06, BP
 	LEAL  4(CX)(BP*1), BP
 	CMPL  BP, 8(SP)
 	JGE   emit_remainder_encodeBlockAsm4MB
+	MOVQ  (DX)(CX*1), SI
 	MOVL  BP, 20(SP)
 	MOVQ  $0x0000cf1bbcdcbf9b, R8
 	MOVQ  SI, R9
@@ -2385,13 +2385,13 @@ zero_loop_encodeBlockAsm12B:
 	MOVQ  src_base+24(FP), DX
 
 search_loop_encodeBlockAsm12B:
-	MOVQ  (DX)(CX*1), SI
 	MOVL  CX, BP
 	SUBL  12(SP), BP
 	SHRL  $0x05, BP
 	LEAL  4(CX)(BP*1), BP
 	CMPL  BP, 8(SP)
 	JGE   emit_remainder_encodeBlockAsm12B
+	MOVQ  (DX)(CX*1), SI
 	MOVL  BP, 20(SP)
 	MOVQ  $0x000000cf1bbcdcbb, R8
 	MOVQ  SI, R9
@@ -3289,13 +3289,13 @@ zero_loop_encodeBlockAsm10B:
 	MOVQ  src_base+24(FP), DX
 
 search_loop_encodeBlockAsm10B:
-	MOVQ  (DX)(CX*1), SI
 	MOVL  CX, BP
 	SUBL  12(SP), BP
 	SHRL  $0x05, BP
 	LEAL  4(CX)(BP*1), BP
 	CMPL  BP, 8(SP)
 	JGE   emit_remainder_encodeBlockAsm10B
+	MOVQ  (DX)(CX*1), SI
 	MOVL  BP, 20(SP)
 	MOVQ  $0x9e3779b1, R8
 	MOVQ  SI, R9
@@ -4193,13 +4193,13 @@ zero_loop_encodeBlockAsm8B:
 	MOVQ  src_base+24(FP), DX
 
 search_loop_encodeBlockAsm8B:
-	MOVQ  (DX)(CX*1), SI
 	MOVL  CX, BP
 	SUBL  12(SP), BP
 	SHRL  $0x04, BP
 	LEAL  4(CX)(BP*1), BP
 	CMPL  BP, 8(SP)
 	JGE   emit_remainder_encodeBlockAsm8B
+	MOVQ  (DX)(CX*1), SI
 	MOVL  BP, 20(SP)
 	MOVQ  $0x9e3779b1, R8
 	MOVQ  SI, R9
@@ -5081,13 +5081,13 @@ zero_loop_encodeBetterBlockAsm:
 	MOVQ  src_base+24(FP), DX
 
 search_loop_encodeBetterBlockAsm:
-	MOVQ  (DX)(CX*1), SI
 	MOVL  CX, BP
 	SUBL  12(SP), BP
 	SHRL  $0x07, BP
 	LEAL  1(CX)(BP*1), BP
 	CMPL  BP, 8(SP)
 	JGE   emit_remainder_encodeBetterBlockAsm
+	MOVQ  (DX)(CX*1), SI
 	MOVL  BP, 20(SP)
 	MOVQ  $0x00cf1bbcdcbfa563, R8
 	MOVQ  $0x9e3779b1, BP
@@ -6059,13 +6059,13 @@ zero_loop_encodeBetterBlockAsm4MB:
 	MOVQ  src_base+24(FP), DX
 
 search_loop_encodeBetterBlockAsm4MB:
-	MOVQ  (DX)(CX*1), SI
 	MOVL  CX, BP
 	SUBL  12(SP), BP
 	SHRL  $0x07, BP
 	LEAL  1(CX)(BP*1), BP
 	CMPL  BP, 8(SP)
 	JGE   emit_remainder_encodeBetterBlockAsm4MB
+	MOVQ  (DX)(CX*1), SI
 	MOVL  BP, 20(SP)
 	MOVQ  $0x00cf1bbcdcbfa563, R8
 	MOVQ  $0x9e3779b1, BP
@@ -6980,13 +6980,13 @@ zero_loop_encodeBetterBlockAsm12B:
 	MOVQ  src_base+24(FP), DX
 
 search_loop_encodeBetterBlockAsm12B:
-	MOVQ  (DX)(CX*1), SI
 	MOVL  CX, BP
 	SUBL  12(SP), BP
 	SHRL  $0x06, BP
 	LEAL  1(CX)(BP*1), BP
 	CMPL  BP, 8(SP)
 	JGE   emit_remainder_encodeBetterBlockAsm12B
+	MOVQ  (DX)(CX*1), SI
 	MOVL  BP, 20(SP)
 	MOVQ  $0x0000cf1bbcdcbf9b, R8
 	MOVQ  $0x9e3779b1, BP
@@ -7755,13 +7755,13 @@ zero_loop_encodeBetterBlockAsm10B:
 	MOVQ  src_base+24(FP), DX
 
 search_loop_encodeBetterBlockAsm10B:
-	MOVQ  (DX)(CX*1), SI
 	MOVL  CX, BP
 	SUBL  12(SP), BP
 	SHRL  $0x05, BP
 	LEAL  1(CX)(BP*1), BP
 	CMPL  BP, 8(SP)
 	JGE   emit_remainder_encodeBetterBlockAsm10B
+	MOVQ  (DX)(CX*1), SI
 	MOVL  BP, 20(SP)
 	MOVQ  $0x0000cf1bbcdcbf9b, R8
 	MOVQ  $0x9e3779b1, BP
@@ -8530,13 +8530,13 @@ zero_loop_encodeBetterBlockAsm8B:
 	MOVQ  src_base+24(FP), DX
 
 search_loop_encodeBetterBlockAsm8B:
-	MOVQ  (DX)(CX*1), SI
 	MOVL  CX, BP
 	SUBL  12(SP), BP
 	SHRL  $0x04, BP
 	LEAL  1(CX)(BP*1), BP
 	CMPL  BP, 8(SP)
 	JGE   emit_remainder_encodeBetterBlockAsm8B
+	MOVQ  (DX)(CX*1), SI
 	MOVL  BP, 20(SP)
 	MOVQ  $0x0000cf1bbcdcbf9b, R8
 	MOVQ  $0x9e3779b1, BP
@@ -9295,13 +9295,13 @@ zero_loop_encodeSnappyBlockAsm:
 	MOVQ  src_base+24(FP), DX
 
 search_loop_encodeSnappyBlockAsm:
-	MOVQ  (DX)(CX*1), SI
 	MOVL  CX, BP
 	SUBL  12(SP), BP
 	SHRL  $0x06, BP
 	LEAL  4(CX)(BP*1), BP
 	CMPL  BP, 8(SP)
 	JGE   emit_remainder_encodeSnappyBlockAsm
+	MOVQ  (DX)(CX*1), SI
 	MOVL  BP, 20(SP)
 	MOVQ  $0x0000cf1bbcdcbf9b, R8
 	MOVQ  SI, R9
@@ -10169,13 +10169,13 @@ zero_loop_encodeSnappyBlockAsm12B:
 	MOVQ  src_base+24(FP), DX
 
 search_loop_encodeSnappyBlockAsm12B:
-	MOVQ  (DX)(CX*1), SI
 	MOVL  CX, BP
 	SUBL  12(SP), BP
 	SHRL  $0x05, BP
 	LEAL  4(CX)(BP*1), BP
 	CMPL  BP, 8(SP)
 	JGE   emit_remainder_encodeSnappyBlockAsm12B
+	MOVQ  (DX)(CX*1), SI
 	MOVL  BP, 20(SP)
 	MOVQ  $0x000000cf1bbcdcbb, R8
 	MOVQ  SI, R9
@@ -10938,13 +10938,13 @@ zero_loop_encodeSnappyBlockAsm10B:
 	MOVQ  src_base+24(FP), DX
 
 search_loop_encodeSnappyBlockAsm10B:
-	MOVQ  (DX)(CX*1), SI
 	MOVL  CX, BP
 	SUBL  12(SP), BP
 	SHRL  $0x05, BP
 	LEAL  4(CX)(BP*1), BP
 	CMPL  BP, 8(SP)
 	JGE   emit_remainder_encodeSnappyBlockAsm10B
+	MOVQ  (DX)(CX*1), SI
 	MOVL  BP, 20(SP)
 	MOVQ  $0x9e3779b1, R8
 	MOVQ  SI, R9
@@ -11707,13 +11707,13 @@ zero_loop_encodeSnappyBlockAsm8B:
 	MOVQ  src_base+24(FP), DX
 
 search_loop_encodeSnappyBlockAsm8B:
-	MOVQ  (DX)(CX*1), SI
 	MOVL  CX, BP
 	SUBL  12(SP), BP
 	SHRL  $0x04, BP
 	LEAL  4(CX)(BP*1), BP
 	CMPL  BP, 8(SP)
 	JGE   emit_remainder_encodeSnappyBlockAsm8B
+	MOVQ  (DX)(CX*1), SI
 	MOVL  BP, 20(SP)
 	MOVQ  $0x9e3779b1, R8
 	MOVQ  SI, R9