GDC-4.8.1 assembler output for dot product function.

avx2dotproduct.asm

	.section	.text._D4simd10dotproduct18__T10dotProductTfZ10dotProductFNexAfxAfZf,"axG",@progbits,_D4simd10dotproduct18__T10dotProductTfZ10dotProductFNexAfxAfZf,comdat
	.p2align 4,,15
	.weak	_D4simd10dotproduct18__T10dotProductTfZ10dotProductFNexAfxAfZf
	.type	_D4simd10dotproduct18__T10dotProductTfZ10dotProductFNexAfxAfZf, @function
_D4simd10dotproduct18__T10dotProductTfZ10dotProductFNexAfxAfZf:
.LFB412:
	.cfi_startproc
	mov	rdx, rdi
	lea	r10, [rsi+rdi*4]
	and	rdi, -32
	lea	rdi, [rsi+rdi*4]
	and	rdx, -8
	mov	rax, rsi
	mov	r8, rcx
	lea	r9, [rsi+rdx*4]
	cmp	rsi, rdi
	jae	.L2454
	vxorps	xmm1, xmm1, xmm1
	mov	r11, rcx
	vmovaps	ymm2, ymm1
	vmovaps	ymm0, ymm1
	vmovaps	ymm3, ymm1
	.p2align 4,,10
	.p2align 3
.L2448:
	vmovups	ymm5, YMMWORD PTR [rax]
	sub	rax, -128
	sub	r11, -128
	vmovups	ymm4, YMMWORD PTR [r11-128]
	vmovups	ymm6, YMMWORD PTR [rax-96]
	vmovups	ymm7, YMMWORD PTR [r11-96]
	vfmadd231ps	ymm3, ymm5, ymm4
	vmovups	ymm8, YMMWORD PTR [rax-64]
	vmovups	ymm9, YMMWORD PTR [r11-64]
	vfmadd231ps	ymm0, ymm6, ymm7
	vmovups	ymm10, YMMWORD PTR [rax-32]
	vmovups	ymm11, YMMWORD PTR [r11-32]
	cmp	rdi, rax
	vfmadd231ps	ymm2, ymm8, ymm9
	vfmadd231ps	ymm1, ymm10, ymm11
	ja	.L2448
	vaddps	ymm12, ymm3, ymm0
	mov	rax, rsi
	not	rax
	vaddps	ymm13, ymm2, ymm1
	add	rdi, rax
	and	rdi, -128
	lea	r8, [rdi+128]
	lea	rax, [rsi+r8]
	add	r8, rcx
	vaddps	ymm14, ymm12, ymm13
.L2446:
	cmp	r9, rax
	jbe	.L2449
	mov	rcx, r8
	mov	rsi, rax
	.p2align 4,,10
	.p2align 3
.L2451:
	vmovups	ymm15, YMMWORD PTR [rsi]
	add	rsi, 32
	add	rcx, 32
	vmovups	ymm1, YMMWORD PTR [rcx-32]
	cmp	r9, rsi
	vfmadd231ps	ymm14, ymm15, ymm1
	ja	.L2451
	mov	rdx, rax
	not	rdx
	add	rdx, r9
	and	rdx, -32
	add	rdx, 32
	add	rax, rdx
	add	r8, rdx
.L2449:
	vhaddps	ymm2, ymm14, ymm14
	cmp	r10, rax
	vhaddps	ymm0, ymm2, ymm2
	vperm2f128	ymm3, ymm0, ymm0, 17
	vaddps	ymm0, ymm0, ymm3
	jbe	.L2455
	.p2align 4,,10
	.p2align 3
.L2453:
	vmovss	xmm5, DWORD PTR [rax]
	add	rax, 4
	vfmadd231ss	xmm0, xmm5, DWORD PTR [r8]
	add	r8, 4
	cmp	r10, rax
	ja	.L2453
.L2455:
	vzeroupper
	ret