PSHUFFLE_BYTE_FLIP_MASK<>(SB), RODATA|NOPTR, $32 // func blockAVX2(dig *Digest, p []byte) // Requires: AVX, AVX2, BMI2 TEXT ·blockAVX2(SB), NOSPLIT, $56-32 MOVQ dig+0(FP), SI MOVQ p_base+8(FP), DI MOVQ p_len+16(FP), DX SHRQ $0x07, DX SHLQ $0x07, DX JZ done_hash ADDQ DI, DX MOVQ DX, 48(SP) MOVQ (SI), AX MOVQ 8(SI), BX MOVQ 16(SI), CX MOVQ 24(SI), R8 MOVQ 32(SI), DX MOVQ 40(SI), R9 MOVQ 48(SI), R10 MOVQ 56(SI), R11 VMOVDQU PSHUFFLE_BYTE_FLIP_MASK<>+0(SB), Y9 loop0: MOVQ $·_K+0(SB), BP VMOVDQU (DI),...

NOT EDIT. //go:build !purego #include "textflag.h" // func blockAVX2(dig *Digest, p []byte) // Requires: AVX, AVX2, BMI2 TEXT ·blockAVX2(SB), NOSPLIT, $56-32 MOVQ dig+0(FP), SI MOVQ p_base+8(FP), DI MOVQ p_len+16(FP), DX SHRQ $0x07, DX SHLQ $0x07, DX JZ done_hash ADDQ DI, DX MOVQ DX, 48(SP) MOVQ (SI), AX MOVQ 8(SI), BX MOVQ 16(SI), CX MOVQ 24(SI), R8 MOVQ 32(SI), DX MOVQ 40(SI), R9 MOVQ 48(SI), R10 MOVQ 56(SI), R11 VMOVDQU PSHUFFLE_BYTE_FLIP_MASK<>+0(SB), Y9 loop0: MOVQ $·_K+0(SB), BP VMOVDQU (DI),...