.text
	.file	"vecadd.cu"
	.globl	_Z6vecAddPdS_S_i                # -- Begin function _Z6vecAddPdS_S_i
	.p2align	4, 0x90
	.type	_Z6vecAddPdS_S_i,@function
_Z6vecAddPdS_S_i:                       # @_Z6vecAddPdS_S_i
.L_Z6vecAddPdS_S_i$local:
# %bb.0:                                # %_after_block_sync_1
	pushq	%rbp
	movq	%rsp, %rbp
	pushq	%r15
	pushq	%r14
	pushq	%r13
	pushq	%r12
	pushq	%rbx
	pushq	%rax
	movl	%ecx, %ebx
	movq	%rdx, %r14
	movq	%rsi, %r15
	movq	%rdi, %r12
	data16
	leaq	block_size@TLSGD(%rip), %rdi
	data16
	data16
	rex64
	callq	__tls_get_addr@PLT
	movl	(%rax), %r13d
	testl	%r13d, %r13d
	je	.LBB0_5
# %bb.1:                                # %_after_block_sync_0.lr.ph
	data16
	leaq	block_index_x@TLSGD(%rip), %rdi
	data16
	data16
	rex64
	callq	__tls_get_addr@PLT
	movq	%rax, -48(%rbp)                 # 8-byte Spill
	data16
	leaq	block_size_x@TLSGD(%rip), %rdi
	data16
	data16
	rex64
	callq	__tls_get_addr@PLT
	movl	(%rax), %esi
	movq	-48(%rbp), %rax                 # 8-byte Reload
	movl	(%rax), %edi
	imull	%esi, %edi
	xorl	%ecx, %ecx
	jmp	.LBB0_2
	.p2align	4, 0x90
.LBB0_4:                                # %intra_warp_inc
                                        #   in Loop: Header=BB0_2 Depth=1
	incl	%ecx
	cmpl	%r13d, %ecx
	jae	.LBB0_5
.LBB0_2:                                # %_after_block_sync_0
                                        # =>This Inner Loop Header: Depth=1
	movl	%ecx, %eax
	cltd
	idivl	%esi
	addl	%edi, %edx
	cmpl	%ebx, %edx
	jge	.LBB0_4
# %bb.3:                                #   in Loop: Header=BB0_2 Depth=1
	movslq	%edx, %rax
	movsd	(%r12,%rax,8), %xmm0            # xmm0 = mem[0],zero
	addsd	(%r15,%rax,8), %xmm0
	movsd	%xmm0, (%r14,%rax,8)
	jmp	.LBB0_4
.LBB0_5:                                # %_after_block_sync_2
	addq	$8, %rsp
	popq	%rbx
	popq	%r12
	popq	%r13
	popq	%r14
	popq	%r15
	popq	%rbp
	retq
.Lfunc_end0:
	.size	_Z6vecAddPdS_S_i, .Lfunc_end0-_Z6vecAddPdS_S_i
                                        # -- End function
	.globl	vecAddPdS_S_i_wrapper           # -- Begin function vecAddPdS_S_i_wrapper
	.p2align	4, 0x90
	.type	vecAddPdS_S_i_wrapper,@function
vecAddPdS_S_i_wrapper:                  # @vecAddPdS_S_i_wrapper
# %bb.0:
	movq	(%rdi), %rax
	movq	8(%rdi), %rcx
	movq	(%rax), %rax
	movq	(%rcx), %rsi
	movq	16(%rdi), %rcx
	movq	(%rcx), %rdx
	movq	24(%rdi), %rcx
	movl	(%rcx), %ecx
	movq	%rax, %rdi
	jmp	.L_Z6vecAddPdS_S_i$local        # TAILCALL
.Lfunc_end1:
	.size	vecAddPdS_S_i_wrapper, .Lfunc_end1-vecAddPdS_S_i_wrapper
                                        # -- End function
	.ident	"Ubuntu clang version 14.0.0-1ubuntu1.1"
	.ident	"clang version 3.8.0 (tags/RELEASE_380/final)"
	.section	".note.GNU-stack","",@progbits