erigon-pulse/consensus/ethash/fnv_hash16_avx2.s

#include "textflag.h"

// func fnvHash16AVX2(data, mix *uint32, prime uint32)
TEXT ·fnvHash16AVX2(SB), NOSPLIT, $0
	MOVQ data+0(FP), AX
	MOVQ mix+8(FP), BX
	MOVL prime+16(FP), X0

	VPBROADCASTD X0, Y0

	VPMULLD (AX), Y0, Y1
	VPXOR (BX), Y1, Y1
	VMOVDQU Y1, (AX)

	VPMULLD 32(AX), Y0, Y0
	VPXOR 32(BX), Y0, Y0
	VMOVDQU Y0, 32(AX)

	VZEROUPPER

	RET
AVX2 fnvHash. Bump numOfGoroutines to 8 in stage_senders.go (#656) * Profile all stages * Try to recover senders with 8 goroutines * fix CPU profiling for stage_bodies * fix out-of-index * Try full DAG for verfication of header seals * Try to unroll fnvHash for performance * SSE2 assembly for fnvHash16 * fnvHash16AVX2 * Revert changes to state.go * check we're on 64-bit in useAVX2 * Shave a move off fnvHash16AVX2 * asmdecl doesn't know about VMOVD * disable linter in the right place 2020-06-12 11:48:59 +00:00			`#include "textflag.h"`

			`// func fnvHash16AVX2(data, mix *uint32, prime uint32)`
			`TEXT ·fnvHash16AVX2(SB), NOSPLIT, $0`
Use MOV instead of VMOV in fnvHash16AVX2 to peacify asmdecl (#699) 2020-06-30 10:23:28 +00:00			`MOVQ data+0(FP), AX`
			`MOVQ mix+8(FP), BX`
			`MOVL prime+16(FP), X0`
AVX2 fnvHash. Bump numOfGoroutines to 8 in stage_senders.go (#656) * Profile all stages * Try to recover senders with 8 goroutines * fix CPU profiling for stage_bodies * fix out-of-index * Try full DAG for verfication of header seals * Try to unroll fnvHash for performance * SSE2 assembly for fnvHash16 * fnvHash16AVX2 * Revert changes to state.go * check we're on 64-bit in useAVX2 * Shave a move off fnvHash16AVX2 * asmdecl doesn't know about VMOVD * disable linter in the right place 2020-06-12 11:48:59 +00:00
			`VPBROADCASTD X0, Y0`

			`VPMULLD (AX), Y0, Y1`
			`VPXOR (BX), Y1, Y1`
			`VMOVDQU Y1, (AX)`

			`VPMULLD 32(AX), Y0, Y0`
			`VPXOR 32(BX), Y0, Y0`
			`VMOVDQU Y0, 32(AX)`

			`VZEROUPPER`

			`RET`