SIMD experiment with Odin

a.odin

package main

import "core:fmt"
import "core:simd"

main :: proc() {
  i := index_any([]u8{'a', 'a', 'a', 'a', 'a', 'a', 'a', 'a', 'a', 'a', 'a', 'a', 'a', '\r', '&', 'a'}, 16)
  fmt.println(i)
}

ch1: u8 = '&'
ch2: u8 = '\r'
lane :: 16
needle1 := simd.from_array([lane]u8{ch1, ch1, ch1, ch1, ch1, ch1, ch1, ch1, ch1, ch1, ch1, ch1, ch1, ch1, ch1, ch1,})
needle2 := simd.from_array([lane]u8{ch2, ch2, ch2, ch2, ch2, ch2, ch2, ch2, ch2, ch2, ch2, ch2, ch2, ch2, ch2, ch2,})

index_any :: proc(haystack: []u8, len: int) -> int {
    offset: int = 0
    for offset + lane <= len {
        v := simd.from_slice(simd.u8x16, haystack[offset:offset+lane])
        eq1 := simd.lanes_eq(v, needle1)
        eq2 := simd.lanes_eq(v, needle2)

        if simd.reduce_or(eq1) == 0xff || simd.reduce_or(eq2) == 0xff {
            break
        }

        offset += lane
    }

    for i in offset..<len {
        if haystack[i] == ch1 || haystack[i] == ch2 {
            return i
        }
    }

    return -1
}

odin-simd-experiment.S

# ...(snip)...

	.p2align	4, 0x90
	.type	main.index_any,@function
main.index_any:
	.cfi_startproc
	subq	$216, %rsp
	.cfi_def_cfa_offset 224
	movq	%rcx, 40(%rsp)
	movq	%rdx, 48(%rsp)
	movq	%rsi, 56(%rsp)
	movq	%rdi, 64(%rsp)
	movq	48(%rsp), %rax
	movq	56(%rsp), %rcx
	movq	64(%rsp), %rdx
	movq	%rdx, 200(%rsp)
	movq	%rcx, 208(%rsp)
	movq	%rax, 192(%rsp)
	movq	$0, 184(%rsp)
.LBB28_2:
	movq	48(%rsp), %rcx
	movq	184(%rsp), %rax
	addq	$16, %rax
	cmpq	%rcx, %rax
	setle	%al
	andb	$1, %al
	cmpb	$0, %al
	je	.LBB28_7
	movq	184(%rsp), %r8
	movq	%r8, 32(%rsp)
	movq	%r8, %r9
	addq	$16, %r9
	movq	208(%rsp), %rcx
	movq	%rsp, %rax
	movq	%rcx, (%rax)
	movl	$.Lcsbs$da, %edi
	movl	$42, %esi
	movl	$20, %edx
	movl	$50, %ecx
	callq	runtime.slice_expr_error_lo_hi
	movq	32(%rsp), %rcx
	movq	40(%rsp), %rdx
	movq	200(%rsp), %rax
	addq	%rcx, %rax
	movq	%rax, 168(%rsp)
	movq	$16, 176(%rsp)
	movq	168(%rsp), %rdi
	movq	176(%rsp), %rsi
	callq	"simd.from_slice-19570"
	movsd	%xmm1, 152(%rsp)
	movq	%xmm0, 144(%rsp)
	movaps	144(%rsp), %xmm0
	movaps	%xmm0, 128(%rsp)
	movaps	128(%rsp), %xmm0
	movaps	main.needle1(%rip), %xmm1
	pcmpeqb	%xmm1, %xmm0
	movaps	%xmm0, 112(%rsp)
	movaps	128(%rsp), %xmm0
	movaps	main.needle2(%rip), %xmm1
	pcmpeqb	%xmm1, %xmm0
	movaps	%xmm0, 96(%rsp)
	movaps	112(%rsp), %xmm0
	pshufd	$238, %xmm0, %xmm1
	por	%xmm1, %xmm0
	pshufd	$85, %xmm0, %xmm1
	por	%xmm1, %xmm0
	movaps	%xmm0, %xmm1
	psrld	$16, %xmm1
	por	%xmm1, %xmm0
	movaps	%xmm0, %xmm1
	psrlw	$8, %xmm1
	por	%xmm1, %xmm0
	movd	%xmm0, %eax
	cmpb	$-1, %al
	sete	%al
	andb	$1, %al
	cmpb	$0, %al
	jne	.LBB28_5
	movaps	96(%rsp), %xmm0
	pshufd	$238, %xmm0, %xmm1
	por	%xmm1, %xmm0
	pshufd	$85, %xmm0, %xmm1
	por	%xmm1, %xmm0
	movaps	%xmm0, %xmm1
	psrld	$16, %xmm1
	por	%xmm1, %xmm0
	movaps	%xmm0, %xmm1
	psrlw	$8, %xmm1
	por	%xmm1, %xmm0
	movd	%xmm0, %eax
	cmpb	$-1, %al
	sete	%al
	andb	$1, %al
	cmpb	$0, %al
	je	.LBB28_6
.LBB28_5:
	jmp	.LBB28_7
.LBB28_6:
	movq	184(%rsp), %rax
	addq	$16, %rax
	movq	%rax, 184(%rsp)
	jmp	.LBB28_2
.LBB28_7:
	movq	184(%rsp), %rax
	movq	%rax, 88(%rsp)
	movq	$0, 80(%rsp)
.LBB28_8:
	movq	48(%rsp), %rax
	cmpq	%rax, 88(%rsp)
	jge	.LBB28_14
	movq	56(%rsp), %r9
	movq	88(%rsp), %rax
	movq	%rax, 72(%rsp)
	movq	72(%rsp), %r8
	movq	%r8, 24(%rsp)
	movl	$.Lcsbs$da, %edi
	movl	$42, %esi
	movl	$32, %edx
	movl	$21, %ecx
	callq	runtime.bounds_check_error
	movq	24(%rsp), %rcx
	movq	64(%rsp), %rax
	movb	(%rax,%rcx), %al
	cmpb	main.ch1, %al
	sete	%al
	andb	$1, %al
	cmpb	$0, %al
	jne	.LBB28_11
	movq	56(%rsp), %r9
	movq	72(%rsp), %r8
	movq	%r8, 16(%rsp)
	movl	$.Lcsbs$da, %edi
	movl	$42, %esi
	movl	$32, %edx
	movl	$43, %ecx
	callq	runtime.bounds_check_error
	movq	16(%rsp), %rcx
	movq	64(%rsp), %rax
	movb	(%rax,%rcx), %al
	cmpb	main.ch2, %al
	sete	%al
	andb	$1, %al
	cmpb	$0, %al
	je	.LBB28_12
.LBB28_11:
	movq	72(%rsp), %rax
	addq	$216, %rsp
	.cfi_def_cfa_offset 8
	retq
.LBB28_12:
	.cfi_def_cfa_offset 224
	jmp	.LBB28_13
.LBB28_13:
	movq	88(%rsp), %rax
	addq	$1, %rax
	movq	%rax, 88(%rsp)
	movq	80(%rsp), %rax
	addq	$1, %rax
	movq	%rax, 80(%rsp)
	jmp	.LBB28_8
.LBB28_14:
	movq	$-1, %rax
	addq	$216, %rsp
	.cfi_def_cfa_offset 8
	retq
.Lfunc_end28:
	.size	main.index_any, .Lfunc_end28-main.index_any
	.cfi_endproc

# ...(snip)...

My attempt to port the example at https://x.com/orisano/status/1813187886910697632 to Odin.

odin-simd-experiment.S is build with the following command:

$ odin build . -build-mode:asm

$ odin version
odin version dev-2024-07-nightly:b4ca044