Page MenuHomePhabricator
Feed Advanced Search

Jul 7 2014

nadav added a comment to D4405: [x86] Add a ZERO_EXTEND_VECTOR_INREG DAG node and use it when widening vector types to be legal and a ZERO_EXTEND node is encountered..

LGTM.

Jul 7 2014, 4:50 PM

Jul 3 2014

nadav added a comment to D4374: [X86] Limit maximum nop length on Silvermont.

LGTM.

Jul 3 2014, 2:22 PM

Jun 5 2014

nadav added a comment to D4039: Fixed a bug in lowering shuffle_vectors to insertps.

LGTM.

Jun 5 2014, 7:58 PM
nadav added a comment to D3800: Fix PR19657 : SLP vectorization doesn't combine scalar load to vector loads.

LGTM. Please commit.

Jun 5 2014, 9:20 AM

May 27 2014

nadav added a comment to D3800: Fix PR19657 : SLP vectorization doesn't combine scalar load to vector loads.

Please document the code that you are adding to the vectorization of binary operators.

May 27 2014, 10:25 PM
nadav added a comment to D3859: Convert some X86 blendv* intrinsics into IR..

Please commit the patch with the fix to the commuted argument order and a fix to the typo in the docs.

May 27 2014, 9:46 AM

May 23 2014

nadav added a comment to D3800: Fix PR19657 : SLP vectorization doesn't combine scalar load to vector loads.

Hi Karthik,

May 23 2014, 9:32 AM

May 22 2014

nadav added a comment to D3851: Add support to vectorize ctlz,cttz and powi intrinsics in SLPVectorizer.

Please add a testcase for one of the functions where the last argument is different and the SLPvectorizer is unable to vectorize the function.

May 22 2014, 9:02 AM

May 21 2014

nadav added a comment to D3863: Do not generate a __morestack prologue, for functions with a stack of size zero.

Hi Florian,

May 21 2014, 9:40 PM

May 20 2014

nadav added a comment to D3851: Add support to vectorize ctlz,cttz and powi intrinsics in SLPVectorizer.

Hi Karthik,

May 20 2014, 11:58 AM

May 19 2014

nadav added a comment to D3800: Fix PR19657 : SLP vectorization doesn't combine scalar load to vector loads.

Hi Eric,

May 19 2014, 11:20 AM
nadav added a comment to D3826: [X86] Tune LEA usage for Silvermont.

+ bool isSafeToClobberEFLAGS(MachineBasicBlock &MBB,
+ MachineBasicBlock::iterator I) const;
+

May 19 2014, 9:29 AM
nadav added a comment to D3821: [LV][REFACTOR] One more tiny fix for printing debug locations in loop vectorizer. Now consistent with the remarks emitter..

LGTM.

May 19 2014, 9:23 AM

May 16 2014

nadav added a comment to D3800: Fix PR19657 : SLP vectorization doesn't combine scalar load to vector loads.

I am sorry but this is the wrong approach. You copy a ton of code that we will need to maintain if we want to add support to new instructions.

May 16 2014, 8:01 AM

May 14 2014

nadav added a comment to D3752: [X86 DAG] Lower vselects with a constant condition as blend+imm if possible.

Excellent. Please update the cost model as well.

May 14 2014, 2:49 PM
nadav added a comment to D3752: [X86 DAG] Lower vselects with a constant condition as blend+imm if possible.

Overall this change makes sense. I did not review the patch carefully but I noticed that Andrea did. When you commit please split this patch to the different optimizations: shuffle blends, BLENDI, etc.

May 14 2014, 1:22 PM

May 9 2014

nadav added a comment to D3525: Reorder shuffle and binary operation..

I did not review the code carefully but I think it looks fine. We are not generating new shuffle masks so there should not be any lowering problems.

May 9 2014, 12:39 PM

May 6 2014

nadav added a comment to D3513: [LV][BUG] Tiny fix for printing debug locations for absolute paths..

LGTM.

May 6 2014, 9:34 AM

May 5 2014

nadav added a comment to D3513: [LV][BUG] Tiny fix for printing debug locations for absolute paths..

Why is getDebugLocString defined in LoopVectorize.cpp? I am expecting other passes to also need this utility. Please refactor this function and put this function in a place where other passes would be able to use it.

May 5 2014, 9:13 AM

May 2 2014

nadav added a comment to D3560: Vectorize math intrinsic calls in SLPVectorizer.

The patch LGTM. Did you run it on the LLVM test suite? Do you know of any workloads that improve?

May 2 2014, 10:23 AM

Apr 29 2014

nadav added a comment to D3559: Move getIntrinsicIDForCall from LoopVectorize to VectorUtil.

This part LGTM.

Apr 29 2014, 10:57 PM
nadav added a comment to D3535: Vectorize intrinsic math function calls in SLPVectorizer.

Hi Karthik,

Apr 29 2014, 9:17 AM

Apr 24 2014

nadav added a comment to D3494: [DAGCombiner] Fold (vextract v, (sext (trunc i))) -> (vextract v, i).

LGTM.

Apr 24 2014, 4:36 PM

Apr 23 2014

nadav added a comment to D3475: Optimization for certain shufflevector by using insertps..

I did not review the patch carefully but from a quick look it looks fine. Andrea, what do you say?

Apr 23 2014, 9:50 PM