[MIPS]Remove unnecessary SLL instructions on MIPS64el #109386

anbbna · 2024-09-20T07:59:40Z

Optimize ((signext (xor (trunc X), imm)) to (xor (X, imm)).

Fix #99783

github-actions · 2024-09-20T08:03:07Z

✅ With the latest revision this PR passed the C/C++ code formatter.

brad0 · 2024-09-22T01:00:50Z

cc @topperc @wzssyqa @yingopq

brad0 · 2024-10-03T23:06:59Z

@nikic

wzssyqa · 2024-10-09T12:42:46Z

llvm/lib/Target/Mips/MipsISelLowering.cpp

+  EVT VT = N->getValueType(0);
+
+  // Pattern match XOR.
+  //  $dst = sign_extend (xor (trunc $src), imm)


trunc needs an argument. I guess the argument will have some effects on the result.

For example, if $src=0x12345678; imm=0xffff.

lets truncate it to i8, then it will be 0x78, the result will be ~0x78.
lets truncate it to i16, then it will be 0x5678, the result will be ~0x5678.

wzssyqa

Maybe it is not correct.

anbbna · 2024-10-12T06:23:30Z

Maybe it is not correct.

Thanks for your review, I would find another solution.

anbbna · 2024-11-26T07:51:31Z

@wzssyqa Could you help review again, thanks!

wzssyqa · 2024-11-26T08:42:56Z

Please read: https://llvm.org/docs/LangRef.html#id2158
The IR instruction trunc has a IMM argument. It will be quilte different if the IMM is different.

yingopq · 2024-12-23T03:42:03Z

llvm/lib/Target/Mips/MipsISelLowering.cpp

+    SDValue Xor = DAG.getNode(ISD::XOR, SDLoc(N), VT, TruncateOperand,
+                              DAG.getTargetConstant(Mask, SDLoc(N), VT));
+    SDValue Truncate =
+        DAG.getNode(ISD::TRUNCATE, SDLoc(N), N0->getValueType(0), Xor);


@wzssyqa The third parameter N0->getValueType(0) is the truncate's destination type, that you mentioned the code lacked. Through this Node we can truncate the fourth parameter Xor to type N0->getValueType(0).
So I think this did not lack a imm. May I understand right?

brad0 · 2025-01-26T05:57:26Z

Ping.

nikic

I don't think this transform is correct: https://alive2.llvm.org/ce/z/-kNt7y

topperc · 2025-01-26T18:56:03Z

llvm/lib/Target/Mips/MipsISelLowering.cpp

+    SDValue TruncateOperand = N0.getOperand(0).getOperand(0);
+    if (VT == MVT::i64 && VT == TruncateOperand->getValueType(0)) {
+      APInt MinusOne(32, -1, true);
+      if (N0.getConstantOperandAPInt(1) == MinusOne) {


N0.getOperandAPInt(1).isAllOnes()?

topperc · 2025-01-26T18:56:41Z

llvm/lib/Target/Mips/MipsISelLowering.cpp

+      APInt MinusOne(32, -1, true);
+      if (N0.getConstantOperandAPInt(1) == MinusOne) {
+        return DAG.getNode(ISD::XOR, SDLoc(N0), VT, TruncateOperand,
+                           DAG.getTargetConstant(-1, SDLoc(N0), VT));


This should not be a target constant. TargetConstant should only nodes that always take an immediate for that operand.

topperc · 2025-01-26T18:57:14Z

llvm/lib/Target/Mips/MipsISelLowering.cpp

+  SDValue N0 = N->getOperand(0);
+  EVT VT = N->getValueType(0);
+
+  if (N0->getNumOperands() != 2) {


This check is unnecessary. The N0.getOpcode() == ISD::XOR check below covers it.

Yes, I will remove it.

topperc · 2025-01-26T19:01:07Z

llvm/lib/Target/Mips/MipsISelLowering.cpp

+      N0.getOperand(0).getOpcode() == ISD::TRUNCATE &&
+      N0.getOperand(1).getOpcode() == ISD::Constant) {
+    SDValue TruncateOperand = N0.getOperand(0).getOperand(0);
+    if (VT == MVT::i64 && VT == TruncateOperand->getValueType(0)) {


This should be TruncateOperand.getValueType(). You don't that result 0 of the TruncateOperand node is being used. The result number is stored in the SDValue so you should use it.

topperc · 2025-01-26T19:03:17Z

I don't think this transform is correct: https://alive2.llvm.org/ce/z/-kNt7y

I agree. You either need to call ComputeNumSignBits on the truncate input or do this transform instead:

((signext (xor (trunc X), imm)) to (xor (signext_inreg X, i32), imm).

anbbna · 2025-01-27T07:00:33Z

I don't think this transform is correct: https://alive2.llvm.org/ce/z/-kNt7y

I agree. You either need to call ComputeNumSignBits on the truncate input or do this transform instead:

((signext (xor (trunc X), imm)) to (xor (signext_inreg X, i32), imm).

Thanks for your review. I will modify it to " (xor (signext_inreg X, i32), imm)" based on your comments.

anbbna · 2025-04-01T08:00:11Z

Ping.

brad0 · 2025-04-12T04:46:49Z

cc @nikic

yingopq · 2025-04-21T08:00:06Z

llvm/test/CodeGen/Mips/llvm-ir/not.ll

-; RUN:    -check-prefixes=ALL,MM,MM32
-; RUN: llc < %s -mtriple=mips -mcpu=mips32r6 -mattr=+micromips | FileCheck %s \
-; RUN:    -check-prefixes=ALL,MM,MM32
+; RUN: llc < %s -march=mips -mcpu=mips2 | FileCheck %s -check-prefixes=GP32


Why change -mtriple to -march?

It was my oversight, I will restore it to -mtriple.

Sorry, I checked the code again and changing -mtriple to -march is the result after running test_checks.py.Because the original code only specifies the CPU architecture .

yingopq · 2025-04-28T07:48:47Z

This new patch LGTM.

llvm/lib/Target/Mips/MipsISelLowering.cpp

llvm/test/CodeGen/Mips/xor-and.ll

…inreg X, i32), imm) The constant operand of the XOR is an i32, which makes the two operands being truncated to i32, and sign extended to i64 afterwards. Since both of the i32 operands are sign extended to i64 before performing the TRUNCATE operation, we just remove the unnecessary TRUNCATE and SIGN_EXTEND. Fix llvm#99783

yingopq · 2025-05-07T07:07:56Z

LGTM.

llvm-ci · 2025-05-07T07:57:10Z

LLVM Buildbot has detected a new failure on builder lldb-x86_64-debian running on lldb-x86_64-debian while building llvm at step 6 "test".

Full details are available at: https://lab.llvm.org/buildbot/#/builders/162/builds/21788

Here is the relevant piece of the build log for the reference

Step 6 (test) failure: build (failure)
...
UNSUPPORTED: lldb-shell :: ScriptInterpreter/Python/Crashlog/interactive_crashlog_json.test (2881 of 2890)
UNSUPPORTED: lldb-shell :: ScriptInterpreter/Lua/watchpoint_callback.test (2882 of 2890)
UNSUPPORTED: lldb-shell :: ScriptInterpreter/Lua/partial_statements.test (2883 of 2890)
PASS: lldb-api :: tools/lldb-server/TestLldbGdbServer.py (2884 of 2890)
PASS: lldb-api :: api/multithreaded/TestMultithreaded.py (2885 of 2890)
PASS: lldb-api :: terminal/TestEditlineCompletions.py (2886 of 2890)
PASS: lldb-api :: commands/process/attach/TestProcessAttach.py (2887 of 2890)
PASS: lldb-api :: repl/clang/TestClangREPL.py (2888 of 2890)
PASS: lldb-api :: tools/lldb-dap/attach/TestDAP_attachByPortNum.py (2889 of 2890)
TIMEOUT: lldb-api :: tools/lldb-dap/breakpoint/TestDAP_setExceptionBreakpoints.py (2890 of 2890)
******************** TEST 'lldb-api :: tools/lldb-dap/breakpoint/TestDAP_setExceptionBreakpoints.py' FAILED ********************
Script:
--
/usr/bin/python3 /home/worker/2.0.1/lldb-x86_64-debian/llvm-project/lldb/test/API/dotest.py -u CXXFLAGS -u CFLAGS --env LLVM_LIBS_DIR=/home/worker/2.0.1/lldb-x86_64-debian/build/./lib --env LLVM_INCLUDE_DIR=/home/worker/2.0.1/lldb-x86_64-debian/build/include --env LLVM_TOOLS_DIR=/home/worker/2.0.1/lldb-x86_64-debian/build/./bin --arch x86_64 --build-dir /home/worker/2.0.1/lldb-x86_64-debian/build/lldb-test-build.noindex --lldb-module-cache-dir /home/worker/2.0.1/lldb-x86_64-debian/build/lldb-test-build.noindex/module-cache-lldb/lldb-api --clang-module-cache-dir /home/worker/2.0.1/lldb-x86_64-debian/build/lldb-test-build.noindex/module-cache-clang/lldb-api --executable /home/worker/2.0.1/lldb-x86_64-debian/build/./bin/lldb --compiler /home/worker/2.0.1/lldb-x86_64-debian/build/./bin/clang --dsymutil /home/worker/2.0.1/lldb-x86_64-debian/build/./bin/dsymutil --make /usr/bin/gmake --llvm-tools-dir /home/worker/2.0.1/lldb-x86_64-debian/build/./bin --lldb-obj-root /home/worker/2.0.1/lldb-x86_64-debian/build/tools/lldb --lldb-libs-dir /home/worker/2.0.1/lldb-x86_64-debian/build/./lib -t /home/worker/2.0.1/lldb-x86_64-debian/llvm-project/lldb/test/API/tools/lldb-dap/breakpoint -p TestDAP_setExceptionBreakpoints.py
--
Exit Code: -9
Timeout: Reached timeout of 600 seconds

Command Output (stdout):
--
lldb version 21.0.0git (https://github.com/llvm/llvm-project.git revision 69f3552d5ee27b4a2fb160600564d3d4d3c3868b)
  clang revision 69f3552d5ee27b4a2fb160600564d3d4d3c3868b
  llvm revision 69f3552d5ee27b4a2fb160600564d3d4d3c3868b

--
Command Output (stderr):
--
Change dir to: /home/worker/2.0.1/lldb-x86_64-debian/llvm-project/lldb/test/API/tools/lldb-dap/breakpoint
runCmd: settings clear --all

output: 

runCmd: settings set symbols.enable-external-lookup false

output: 

runCmd: settings set target.inherit-tcc true

output: 

runCmd: settings set target.disable-aslr false

output: 

runCmd: settings set target.detach-on-error false

output: 

runCmd: settings set target.auto-apply-fixits false

Optimize ((signext (xor (trunc X), imm)) to (xor (X, imm)). Fix llvm#99783

anbbna force-pushed the mips-xor branch 3 times, most recently from 7e8a482 to a3aa217 Compare September 20, 2024 09:32

anbbna force-pushed the mips-xor branch from a3aa217 to a875a05 Compare September 30, 2024 08:36

anbbna force-pushed the mips-xor branch from a875a05 to 8ead8c5 Compare October 9, 2024 08:02

nikic requested a review from topperc October 9, 2024 09:54

wzssyqa reviewed Oct 9, 2024

View reviewed changes

wzssyqa requested changes Oct 9, 2024

View reviewed changes

anbbna force-pushed the mips-xor branch 4 times, most recently from 591d423 to 1fdb6f2 Compare October 29, 2024 10:47

yingopq reviewed Dec 23, 2024

View reviewed changes

anbbna force-pushed the mips-xor branch 2 times, most recently from 525bdfb to 013a7b7 Compare January 26, 2025 11:32

nikic requested changes Jan 26, 2025

View reviewed changes

topperc reviewed Jan 26, 2025

View reviewed changes

anbbna force-pushed the mips-xor branch 2 times, most recently from 046f4d4 to 87852ae Compare February 18, 2025 11:02

brad0 requested a review from wzssyqa April 6, 2025 10:09

brad0 requested a review from topperc April 6, 2025 10:09

brad0 requested review from nikic and yingopq April 20, 2025 04:48

yingopq reviewed Apr 21, 2025

View reviewed changes

brad0 requested a review from yingopq April 27, 2025 07:18

anbbna force-pushed the mips-xor branch from 87852ae to 3b122b9 Compare April 28, 2025 02:45

yingopq reviewed Apr 30, 2025

View reviewed changes

llvm/lib/Target/Mips/MipsISelLowering.cpp Outdated Show resolved Hide resolved

yingopq reviewed Apr 30, 2025

View reviewed changes

llvm/test/CodeGen/Mips/xor-and.ll Show resolved Hide resolved

anbbna force-pushed the mips-xor branch from 3b122b9 to fd2ff11 Compare May 7, 2025 03:34

anbbna force-pushed the mips-xor branch from fd2ff11 to 88c7b1e Compare May 7, 2025 03:42

brad0 merged commit 69f3552 into llvm:main May 7, 2025
8 of 11 checks passed

GeorgeARM pushed a commit to GeorgeARM/llvm-project that referenced this pull request May 7, 2025

[MIPS]Remove unnecessary SLL instructions on MIPS64el (llvm#109386)

65c4d15

Optimize ((signext (xor (trunc X), imm)) to (xor (X, imm)). Fix llvm#99783

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[MIPS]Remove unnecessary SLL instructions on MIPS64el #109386

[MIPS]Remove unnecessary SLL instructions on MIPS64el #109386

anbbna commented Sep 20, 2024 •

edited by wzssyqa

Loading

github-actions bot commented Sep 20, 2024 •

edited

Loading

brad0 commented Sep 22, 2024

brad0 commented Oct 3, 2024

wzssyqa Oct 9, 2024

wzssyqa Oct 9, 2024

wzssyqa left a comment

anbbna commented Oct 12, 2024

anbbna commented Nov 26, 2024

wzssyqa commented Nov 26, 2024

yingopq Dec 23, 2024

brad0 commented Jan 26, 2025

nikic left a comment

topperc Jan 26, 2025

topperc Jan 26, 2025

topperc Jan 26, 2025

anbbna Jan 27, 2025

topperc Jan 26, 2025

topperc commented Jan 26, 2025

anbbna commented Jan 27, 2025

anbbna commented Apr 1, 2025

brad0 commented Apr 12, 2025

yingopq Apr 21, 2025

anbbna Apr 21, 2025

anbbna Apr 25, 2025

yingopq commented Apr 28, 2025

yingopq commented May 7, 2025

llvm-ci commented May 7, 2025

[MIPS]Remove unnecessary SLL instructions on MIPS64el #109386

[MIPS]Remove unnecessary SLL instructions on MIPS64el #109386

Conversation

anbbna commented Sep 20, 2024 • edited by wzssyqa Loading

github-actions bot commented Sep 20, 2024 • edited Loading

brad0 commented Sep 22, 2024

brad0 commented Oct 3, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

wzssyqa left a comment

Choose a reason for hiding this comment

anbbna commented Oct 12, 2024

anbbna commented Nov 26, 2024

wzssyqa commented Nov 26, 2024

Choose a reason for hiding this comment

brad0 commented Jan 26, 2025

nikic left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

topperc commented Jan 26, 2025

anbbna commented Jan 27, 2025

anbbna commented Apr 1, 2025

brad0 commented Apr 12, 2025

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

yingopq commented Apr 28, 2025

yingopq commented May 7, 2025

llvm-ci commented May 7, 2025

anbbna commented Sep 20, 2024 •

edited by wzssyqa

Loading

github-actions bot commented Sep 20, 2024 •

edited

Loading