Optimize 64-bit shifts by constants for x86-32

Optimize 64-bit shifts by constants for x86-32
Hopefully improves perf in fpclassifyd in ammp spec test.

BUG=none
R=stichnot@chromium.org

Committed: https://gerrit.chromium.org/gerrit/gitweb?p=native_client/pnacl-subzero.git;a=commit;h=55f2e6d308e8a05dad706a6dd9c18a8f103e63c8

[sehr, 2015-09-21 23:52:22]

sehr@chromium.org changed reviewers:
+ jpp@chromium.org, stichnot@chromium.org

[sehr, 2015-09-21 23:53:01]

Checkpoint upload before removing commonality.  PTAL.

[Jim Stichnoth, 2015-09-22 05:48:46]

Since you pointed out offline that we aren't actually testing i64 shifts by
values between 32..63, can you add those to the cross tests?

diff --git a/crosstest/test_arith.def b/crosstest/test_arith.def
index 488079d..698f30b 100644
--- a/crosstest/test_arith.def
+++ b/crosstest/test_arith.def
@@ -58,6 +58,7 @@
 #define INT_VALUE_ARRAY                           \
 { 0x0,        0x1,        0x7ffffffe, 0x7fffffff, \
   0x80000000, 0x80000001, 0xfffffffe, 0xffffffff, \
+  0x1e, 0x1f, 0x20, 0x21, 0x3e, 0x3f, 0x40, 0x41, \
   0x7e,       0x7f,       0x80,       0x81,       \
   0xfe,       0xff,       0x100,      0x101,      \
   0x7ffe,     0x7fff,     0x8000,     0x8001,     \

and also run "make format" to get clang-format-approved formatting.

https://codereview.chromium.org/1351133003/diff/1/src/IceTargetLoweringX86Bas...
File src/IceTargetLoweringX86BaseImpl.h (left):

https://codereview.chromium.org/1351133003/diff/1/src/IceTargetLoweringX86Bas...
src/IceTargetLoweringX86BaseImpl.h:1187: // TODO: Refactor the similarities
between Shl, Lshr, and Ashr.
Don't delete this TODO unless you're taking care of it. :)

https://codereview.chromium.org/1351133003/diff/1/src/IceTargetLoweringX86Bas...
File src/IceTargetLoweringX86BaseImpl.h (right):

https://codereview.chromium.org/1351133003/diff/1/src/IceTargetLoweringX86Bas...
src/IceTargetLoweringX86BaseImpl.h:1199: _shl(T_2,
Ctx->getConstantInt32(ShiftAmount-32));
Please run "make -f Makefile.standalone format" to get auto-formatting by
clang-format.

https://codereview.chromium.org/1351133003/diff/1/tests_lit/llvm2ice_tests/sh...
File tests_lit/llvm2ice_tests/shift.ll (right):

https://codereview.chromium.org/1351133003/diff/1/tests_lit/llvm2ice_tests/sh...
tests_lit/llvm2ice_tests/shift.ll:217: 
Remove "trailing whitespace" blank line to placate git.

[sehr, 2015-09-22 16:03:17]

Fixes in.  New CL uploaded.  PTAL.

https://codereview.chromium.org/1351133003/diff/1/src/IceTargetLoweringX86Bas...
File src/IceTargetLoweringX86BaseImpl.h (left):

https://codereview.chromium.org/1351133003/diff/1/src/IceTargetLoweringX86Bas...
src/IceTargetLoweringX86BaseImpl.h:1187: // TODO: Refactor the similarities
between Shl, Lshr, and Ashr.
On 2015/09/22 05:48:46, stichnot wrote:
> Don't delete this TODO unless you're taking care of it. :)

Done.

https://codereview.chromium.org/1351133003/diff/1/src/IceTargetLoweringX86Bas...
File src/IceTargetLoweringX86BaseImpl.h (right):

https://codereview.chromium.org/1351133003/diff/1/src/IceTargetLoweringX86Bas...
src/IceTargetLoweringX86BaseImpl.h:1199: _shl(T_2,
Ctx->getConstantInt32(ShiftAmount-32));
On 2015/09/22 05:48:46, stichnot wrote:
> Please run "make -f Makefile.standalone format" to get auto-formatting by
> clang-format.

Done.

https://codereview.chromium.org/1351133003/diff/1/tests_lit/llvm2ice_tests/sh...
File tests_lit/llvm2ice_tests/shift.ll (right):

https://codereview.chromium.org/1351133003/diff/1/tests_lit/llvm2ice_tests/sh...
tests_lit/llvm2ice_tests/shift.ll:217: 
On 2015/09/22 05:48:46, stichnot wrote:
> Remove "trailing whitespace" blank line to placate git.

Done.

[Jim Stichnoth, 2015-09-22 16:07:41]

lgtm

[sehr, 2015-09-22 16:17:44]

Committed patchset #2 (id:20001) manually as
55f2e6d308e8a05dad706a6dd9c18a8f103e63c8 (presubmit successful).

[John, 2015-09-22 16:18:37]

lgtm

just curious. LGTM otherwise.

https://codereview.chromium.org/1351133003/diff/20001/src/IceTargetLoweringX8...
File src/IceTargetLoweringX86BaseImpl.h (right):

https://codereview.chromium.org/1351133003/diff/20001/src/IceTargetLoweringX8...
src/IceTargetLoweringX86BaseImpl.h:1250: _test(T_1, BitTest);
would _cmp be safer here? (e.g., T_1 == 512)