diff --git a/blake2s/blake2s_386.s b/blake2s/blake2s_386.s
index b01d66b..8a80761 100644
--- a/blake2s/blake2s_386.s
+++ b/blake2s/blake2s_386.s
@@ -290,7 +290,7 @@ GLOBL counter<>(SB), (NOPTR+RODATA), $16
 	MOVL t, 8*4+off+576(dst)
 
 // func hashBlocksSSE2(h *[8]uint32, c *[2]uint32, flag uint32, blocks []byte)
-TEXT ·hashBlocksSSE2(SB), 4, $0-24
+TEXT ·hashBlocksSSE2(SB), 4, $672-24 // frame = 656 + 16 byte alignment
 	MOVL h+0(FP), AX
 	MOVL c+4(FP), BX
 	MOVL flag+8(FP), CX
@@ -298,8 +298,10 @@ TEXT ·hashBlocksSSE2(SB), 4, $0-24
 	MOVL blocks_len+16(FP), DX
 
 	MOVL SP, BP
-	ANDL $0xFFFFFFF0, SP
-	SUBL $(16+16+640), SP
+	MOVL SP, DI
+	ADDL $15, DI
+	ANDL $~15, DI
+	MOVL DI, SP
 
 	MOVL CX, 8(SP)
 	MOVL 0(BX), CX
@@ -357,7 +359,7 @@ loop:
 	RET
 
 // func hashBlocksSSSE3(h *[8]uint32, c *[2]uint32, flag uint32, blocks []byte)
-TEXT ·hashBlocksSSSE3(SB), 4, $0-24
+TEXT ·hashBlocksSSSE3(SB), 4, $704-24 // frame = 688 + 16 byte alignment
 	MOVL h+0(FP), AX
 	MOVL c+4(FP), BX
 	MOVL flag+8(FP), CX
@@ -365,8 +367,10 @@ TEXT ·hashBlocksSSSE3(SB), 4, $0-24
 	MOVL blocks_len+16(FP), DX
 
 	MOVL SP, BP
-	ANDL $0xFFFFFFF0, SP
-	SUBL $(16+16+640+32), SP
+	MOVL SP, DI
+	ADDL $15, DI
+	ANDL $~15, DI
+	MOVL DI, SP
 
 	MOVL CX, 8(SP)
 	MOVL 0(BX), CX
diff --git a/blake2s/blake2s_amd64.s b/blake2s/blake2s_amd64.s
index 7cce0dc..bbecdd3 100644
--- a/blake2s/blake2s_amd64.s
+++ b/blake2s/blake2s_amd64.s
@@ -364,7 +364,7 @@ GLOBL counter<>(SB), (NOPTR+RODATA), $16
 	LOAD_MSG_SSE4(X8, X9, X10, X11, SI, 10, 8, 7, 1, 2, 4, 6, 5, 15, 9, 3, 13, 11, 14, 12, 0); \
 	ROUND_SSSE3(X4, X5, X6, X7, X8, X9, X10, X11, X8, X13, X14)
 
-#define HASH_BLOCKS(h, c, flag, blocks_base, blocks_len, stack_size, BLAKE2s_FUNC) \
+#define HASH_BLOCKS(h, c, flag, blocks_base, blocks_len, BLAKE2s_FUNC) \
 	MOVQ  h, AX;                   \
 	MOVQ  c, BX;                   \
 	MOVL  flag, CX;                \
@@ -372,8 +372,10 @@ GLOBL counter<>(SB), (NOPTR+RODATA), $16
 	MOVQ  blocks_len, DX;          \
 	                               \
 	MOVQ  SP, BP;                  \
-	ANDQ  $0xFFFFFFFFFFFFFFF0, SP; \
-	SUBQ  $(16+16+stack_size), SP; \
+	MOVQ  SP, R9;                  \
+	ADDQ  $15, R9;                 \
+	ANDQ  $~15, R9;                \
+	MOVQ  R9, SP;                  \
 	                               \
 	MOVQ  0(BX), R9;               \
 	MOVQ  R9, 0(SP);               \
@@ -421,18 +423,18 @@ GLOBL counter<>(SB), (NOPTR+RODATA), $16
 	MOVQ  BP, SP
 
 // func hashBlocksSSE2(h *[8]uint32, c *[2]uint32, flag uint32, blocks []byte)
-TEXT ·hashBlocksSSE2(SB), 4, $0-48
-	HASH_BLOCKS(h+0(FP), c+8(FP), flag+16(FP), blocks_base+24(FP), blocks_len+32(FP), 640, BLAKE2s_SSE2)
+TEXT ·hashBlocksSSE2(SB), 4, $672-48 // frame = 656 + 16 byte alignment
+	HASH_BLOCKS(h+0(FP), c+8(FP), flag+16(FP), blocks_base+24(FP), blocks_len+32(FP), BLAKE2s_SSE2)
 	RET
 
 // func hashBlocksSSSE3(h *[8]uint32, c *[2]uint32, flag uint32, blocks []byte)
-TEXT ·hashBlocksSSSE3(SB), 4, $0-48
-	HASH_BLOCKS(h+0(FP), c+8(FP), flag+16(FP), blocks_base+24(FP), blocks_len+32(FP), 640, BLAKE2s_SSSE3)
+TEXT ·hashBlocksSSSE3(SB), 4, $672-48 // frame = 656 + 16 byte alignment
+	HASH_BLOCKS(h+0(FP), c+8(FP), flag+16(FP), blocks_base+24(FP), blocks_len+32(FP), BLAKE2s_SSSE3)
 	RET
 
 // func hashBlocksSSE4(h *[8]uint32, c *[2]uint32, flag uint32, blocks []byte)
-TEXT ·hashBlocksSSE4(SB), 4, $0-48
-	HASH_BLOCKS(h+0(FP), c+8(FP), flag+16(FP), blocks_base+24(FP), blocks_len+32(FP), 0, BLAKE2s_SSE4)
+TEXT ·hashBlocksSSE4(SB), 4, $16-48 // frame = 0 + 16 byte alignment
+	HASH_BLOCKS(h+0(FP), c+8(FP), flag+16(FP), blocks_base+24(FP), blocks_len+32(FP), BLAKE2s_SSE4)
 	RET
 
 // func supportSSE4() bool