This patch looks fine to me.
2011/6/24 Berg, Michael :
> All: The attached patch files are:
>
>
>
> 1.) Correctness changes for cvt operations
>
> 2.) BD translation mapping updates for a set of instructions(AVX)
>
> 3.) Scheduling additions for new spelling of ordered cmp instructions
>
sorry, I meant to get to it, but got bogged down. Give me a couple of days, pls
Sun
On Fri, Jul 1, 2011 at 6:29 AM, Berg, Michael wrote:
> It’s been a while since this request went out, can a gate keeper please take
> look.
>
>
>
> Thx,
>
>
>
> m
>
>
>
> From: Berg, Michael [mailto:michael.b...@a
It's been a while since this request went out, can a gate keeper please take
look.
Thx,
m
From: Berg, Michael [mailto:michael.b...@amd.com]
Sent: Thursday, June 23, 2011 5:58 PM
To: open64-devel@lists.sourceforge.net
Subject: [Open64-devel] Gatekeeper code review request for AMD 4.2.5 merge
Al
In 1.p, the code to check if rtype is complex can be CSEed and move to the
beginning of the block.
2011/6/18 Berg, Michael
> All: The attached are patch files for complex math wrt fma on BD as well
> as some mapping corrections for shift right and movddup.
>
>
>
>
> These changes pa
There are format issues in patch. The original code uses tabs. Could you
please change all the 5 lines to space? Thank you very much.
2011/6/18 Berg, Michael
> All: The attached a single file update for avx operand definitions for
> vector shuffle instructions on x86.
>
>
>
>
> These
Looks good to me.
2011/6/18 Berg, Michael
> All: The attached file enables flag for support to toggle emit of the
> vzeroupper instruction on BD. In some instances performance enhances by
> not emitting the instructions, we currently emit the instruction by default
> to handle avoiding merge
Looks good to me.
2011/6/18 Berg, Michael
> All: The attached file is a simple checkin from our BD work(2 files
> modified). The update extends some support for imac and round instructions
> and updates schedule info for BD.
>
> ** **
>
> These changes pass the following:
>
> a.)
All: The attached a single file update for avx operand definitions for vector
shuffle instructions on x86.
These changes pass the following:
a.) No compile time failure for x86 build.
b.) The gcc regression test suite on x86/Linux with no new failures.
c.) The SPEC20
All: The attached file enables flag for support to toggle emit of the
vzeroupper instruction on BD. In some instances performance enhances by not
emitting the instructions, we currently emit the instruction by default to
handle avoiding merge conflicts in the upper 128bits of ymm instructions
In 2.p, two conditions to decrease the P_i are not consistent:
In CG_LOOP::Determine_Best_Unit_Iteration_Interval():
+#ifdef TARG_X8664
+ // now adjust the number of gpr regs as per the ABI
+ P_i--;
+ if (Is_Target_32bit() && Gen_Frame_Pointer)
+P_i--;
+#endif
In Report_Loop_Info():
+#ifdef
It's been a bit, can a gate keep please review these patches.
Thx,
From: Berg, Michael
Sent: Monday, May 23, 2011 8:45 PM
To: open64-devel@lists.sourceforge.net
Subject: RE: Gatekeeper code review request for AMD 4.2.5 merge
Broken up into much smaller pieces, same request. The last patch file
On Tue, May 24, 2011 at 10:45 AM, Berg, Michael wrote:
> Broken up into much smaller pieces, same request. The last patch file
> contains the bulk of the change and is only property changes.
Much easier to read / thank you very much. I'll forward it to our
engineers for review.
I'm curious abo
On Tue, May 24, 2011 at 9:39 AM, Berg, Michael wrote:
> All: The attached file is a single checkin from our BD branch, which are
> updates to in-flight work on existing code for that target.
>
> The bulk of the changes are to a local late cg scheduler and for BD
> properties in the machin
Berg, Michael wrote:
> Suneel asked for some documentation changes, which are now attached. See the
> patch and the design doc.
> BTW, I still need the review, so if a gatekeeper can review the patch and
> comment I can proceed.
>
If this is 3 things you need to break it up into 3 patches - I
Why does the patch have
+#ifdef KEY
- It may be wise to break any PathScale portions into a separate and
clean commit even if it's small and may break something. Sorry I can't
provide a deeper review.
--
Benefiting fr
15 matches
Mail list logo
