[LLVMdev] SIMD for sdiv <2 x i64> (original) (raw)

zhi chen zchenhn at gmail.com
Fri Jul 24 11:46:43 PDT 2015

Thanks for your help, Philip. The shufflevector could be replaced with insertelement, but I think that shouldn't improve the performance much. I think shufflevector would generate a broadcast or a movlhps/d instruction. The performance bottleneck here is because AVX2 doesn't have a SIMD division for integers. It needs to puck and unpack the vector register and use many shift instructions. It also might use two int division instructions.

Thanks, Zhi

On Fri, Jul 24, 2015 at 11:32 AM, Philip Reames <listmail at philipreames.com> wrote:

This snippet of IR is interesting: %sub.ptr.div.iS37D = sdiv <2 x i64> %sub.ptr.sub.iS36D, <i64 24, i64_ _24> %cmp10S38D = icmp ugt <2 x i64> %sub.ptr.div.iS37D, %splatInsMapS1D.splat %zextS39D = sext <2 x i1> %cmp10S38D to <2 x i64> %BCS39D = bitcast <2 x i64> %zextS39D to i128 %mskS39D = icmp ne i128 %BCS39D, 0 br i1 %mskS39D, label %if.then11, label %if.else

It looks like %msk539D is basically a test of whether either of the vector elements produced by the divide are ugt the spatInstMap. I can't say for sure that we can do better here - I haven't studied our vector canonicalization rules enough - but this seems like something which could possibly be improved. This is interesting: %splatCallS27D.splatinsert = insertelement <2 x i8*> undef, i8* %call5.i.i, i32 0 %splatCallS27D.splat = shufflevector <2 x i8*> %splatCallS27D.splatinsert, <2 x i8*> undef, <2 x i32> zeroinitializer Can't that shuifflevector be replaced with: %splatCallS27D.splat = insertelement <2 x i8*> %splatCallS27D.splatinsert , i8* %call5.i.i, i32 1 Again, without knowledge of how we canonicalize such things, not necessarily a win. Just suspicious. The bitcast/extractelement sequence following that shufflevector is also interesting. It looks like it could be rewritten in terms of the i8* %call5.i.i and a bitcast.

On 07/24/2015 10:52 AM, zhi chen wrote: ------------------------------------ IR ------------------------------------------------------------------ if.then.i.i.i.i.i.i: ; preds = %if.then4 %S25D = zext <2 x i32> %splatLDS17D.splat to <2 x i64> %umulwithoverflow.i.iS26D = shl <2 x i64> %S25D, <i64 3, i64 3> %extumulwithoverflow.i.iS26D = extractelement <2 x i64> %umulwithoverflow.i.iS26D, i32 1 %call5.i.i = tail call noalias i8* @Znam(i64 %extumulwithoverflow.i.iS26D) #22 %splatCallS27D.splatinsert = insertelement <2 x i8*> undef, i8* %call5.i.i, i32 0 %splatCallS27D.splat = shufflevector <2 x i8*> %splatCallS27D.splatinsert, <2 x i8*> undef, <2 x i32> zeroinitializer %bitcastS28D = bitcast <2 x i8*> %splatCallS27D.splat to <2 x double*> %extractS29D = extractelement <2 x double*> %bitcastS28D, i32 1 store double* %extractS29D, double** %val.i.i, align 8 _%val.i3.i.i = getelementptr inbounds %class.Vector* %x, i64 0, i32 3 %4 = load double** %val.i3.i.i, align 8, !tbaa !22 %splatLDS31D.splatinsert = insertelement <2 x double*> undef, double* %4, i32 0 %splatLDS31D.splat = shufflevector <2 x double*> %splatLDS31D.splatinsert, <2 x double*> undef, <2 x i32> zeroinitializer %bitcastS32D = bitcast <2 x double*> %splatLDS31D.splat to <2 x i8*> %extbitcastS32D = extractelement <2 x i8*> %bitcastS32D, i32 1 tail call void @llvm.memmove.p0i8.p0i8.i64(i8* %call5.i.i, i8* %extbitcastS32D, i64 %extumulwithoverflow.i.iS26D, i32 8, i1 false) #9 br label %invoke.cont invoke.cont: ; preds = %if.then.i.i.i.i.i.i, %if.then4 _%sub.ptr.rhs.cast.i = ptrtoint %class.Vector* %position.coerce to i64 %sub.ptr.rhs.cast.iS35D = ptrtoint <2 x %class.Vector*> %splatInsMapS35D.splat to <2 x i64> %sub.ptr.sub.iS36D = sub <2 x i64> %sub.ptr.rhs.castS8D, %sub.ptr.rhs.cast.iS35D %sub.ptr.div.iS37D = sdiv <2 x i64> %sub.ptr.sub.iS36D, <i64 24, i64_ _24> %extractS196D = extractelement <2 x i64> %sub.ptr.div.iS37D, i32 1 %cmp10S38D = icmp ugt <2 x i64> %sub.ptr.div.iS37D, %splatInsMapS1D.splat %zextS39D = sext <2 x i1> %cmp10S38D to <2 x i64> %BCS39D = bitcast <2 x i64> %zextS39D to i128 %mskS39D = icmp ne i128 %BCS39D, 0 br i1 %mskS39D, label %if.then11, label %if.else -------------------------------------------- Assembly ----------------------------------------------------------------- # BB#3: # %if.then.i.i.i.i.i.i vpsllq $3, %xmm0, %xmm0 vpextrq $1, %xmm0, %rbx movq %rbx, %rdi vmovaps %xmm2, 96(%rsp) # 16-byte Spill vmovaps %xmm5, 64(%rsp) # 16-byte Spill vmovdqa %xmm6, 16(%rsp) # 16-byte Spill callq Znam movq %rax, 128(%rsp) movq 16(%r12), %rsi movq %rax, %rdi movq %rbx, %rdx callq memmove vmovdqa 16(%rsp), %xmm6 # 16-byte Reload vmovaps 64(%rsp), %xmm5 # 16-byte Reload vmovaps 96(%rsp), %xmm2 # 16-byte Reload vmovdqa .LCPI5820(%rip), %xmm4 .LBB5824: # %invoke.cont vmovaps %xmm2, 96(%rsp) # 16-byte Spill vmovdqa 48(%rsp), %xmm0 # 16-byte Reload vpsubq %xmm0, %xmm2, %xmm0 vpextrq $1, %xmm0, %rax movabsq $3074457345618258603, %rcx # imm = 0x2AAAAAAAAAAAAAAB imulq %rcx movq %rdx, %rax shrq $63, %rax sarq $2, %rdx addq %rax, %rdx vmovq %rdx, %xmm1 vmovq %xmm0, %rax imulq %rcx movq %rdx, %rax shrq $63, %rax sarq $2, %rdx addq %rax, %rdx vmovq %rdx, %xmm0 vpunpcklqdq %xmm1, %xmm0, %xmm1 # xmm1 = xmm0[0],xmm1[0] vpxor %xmm4, %xmm1, %xmm0 vpcmpgtq %xmm6, %xmm0, %xmm0 vptest %xmm0, %xmm0 je .LBB58249 Thanks, Zhi On Fri, Jul 24, 2015 at 10:16 AM, Philip Reames <listmail at philipreames.com_ _> wrote:

On 07/24/2015 03:42 AM, Benjamin Kramer wrote: On 24.07.2015, at 08:06, zhi chen <zchenhn at gmail.com> wrote:

It seems that that it's hard to vectorize int64 in LLVM. For example, LLVM 3.4 generates very complicated code for the following IR. I am running on a Haswell processor. Is it because there is no alternative AVX/2 instructions for int64? The same thing also happens to zext <2 x i32> -> <2_ _x i64> and trunc <2 x i64> -> <2 x i32>. Any ideas to optimize these instructions? Thanks. %sub.ptr.sub.i6.i.i.i.i = sub <2 x i64> %sub.ptr.lhs.cast.i4.i.i.i.i, %sub.ptr.rhs.cast.i5.i.i.i.i %sub.ptr.div.i7.i.i.i.i = sdiv <2 x i64> %sub.ptr.sub.i6.i.i.i.i, <i64_ _24, i64 24> Assembly: vpsubq %xmm6, %xmm5, %xmm5 vmovq %xmm5, %rax movabsq $3074457345618258603, %rbx # imm = 0x2AAAAAAAAAAAAAAB imulq %rbx movq %rdx, %rcx movq %rcx, %rax shrq $63, %rax shrq $2, %rcx addl %eax, %ecx vpextrq $1, %xmm5, %rax imulq %rbx movq %rdx, %rax shrq $63, %rax shrq $2, %rdx addl %eax, %edx movslq %edx, %rax vmovq %rax, %xmm5 movslq %ecx, %rax vmovq %rax, %xmm6 vpunpcklqdq %xmm5, %xmm6, %xmm5 # xmm5 = xmm6[0],xmm5[0] AVX2 doesn't have integer vector division instructions and LLVM lowers divides by constants into (128 bit) multiplies. However, AVX2 doesn't have a way to get to the upper 64 bits of a 64x64->128 bit multiply either, so LLVM uses the scalar imulq instruction to do that. There's not much room to optimize here given the limitations of AVX2. You seem to be subtracting pointers though, so if you can guarantee that the pointers are aligned you could set the exact bit on your 'sdiv' instruction. That should give better code. Depending on what you're using the result of the divide for, there might be optimizations which could be applied as well. Can you give a slightly larger context for your source IR? (1-2 level of uses/defs out from the instructions would help) - Ben

LLVM Developers mailing list LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20150724/c4bd911c/attachment.html>

More information about the llvm-dev mailing list