maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/X86/unfold-masked-merge-scalar-...

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

467 lines
16 KiB
LLVM
Raw Normal View History

[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=-bmi < %s | FileCheck %s --check-prefix=CHECK-NOBMI
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi < %s | FileCheck %s --check-prefix=CHECK-BMI
; https://bugs.llvm.org/show_bug.cgi?id=37104
; X: [bit 3210]
; Y: [bit 7654]
define i8 @out8_constmask(i8 %x, i8 %y) {
; CHECK-NOBMI-LABEL: out8_constmask:
; CHECK-NOBMI: # %bb.0:
[X86] Enable 8-bit OR with disjoint bits to convert to LEA We already support 8-bits adds in convertToThreeAddress. But we can also support 8-bit OR if the bits are disjoint. We already do this for 16/32/64. Differential Revision: https://reviews.llvm.org/D58863 llvm-svn: 355423
2019-03-06 02:37:33 +08:00
; CHECK-NOBMI-NEXT: # kill: def $esi killed $esi def $rsi
; CHECK-NOBMI-NEXT: # kill: def $edi killed $edi def $rdi
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-NOBMI-NEXT: andb $15, %dil
[X86] Enable 8-bit OR with disjoint bits to convert to LEA We already support 8-bits adds in convertToThreeAddress. But we can also support 8-bit OR if the bits are disjoint. We already do this for 16/32/64. Differential Revision: https://reviews.llvm.org/D58863 llvm-svn: 355423
2019-03-06 02:37:33 +08:00
; CHECK-NOBMI-NEXT: andb $-16, %sil
; CHECK-NOBMI-NEXT: leal (%rsi,%rdi), %eax
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-NOBMI-NEXT: # kill: def $al killed $al killed $eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-NOBMI-NEXT: retq
;
; CHECK-BMI-LABEL: out8_constmask:
; CHECK-BMI: # %bb.0:
[X86] Enable 8-bit OR with disjoint bits to convert to LEA We already support 8-bits adds in convertToThreeAddress. But we can also support 8-bit OR if the bits are disjoint. We already do this for 16/32/64. Differential Revision: https://reviews.llvm.org/D58863 llvm-svn: 355423
2019-03-06 02:37:33 +08:00
; CHECK-BMI-NEXT: # kill: def $esi killed $esi def $rsi
; CHECK-BMI-NEXT: # kill: def $edi killed $edi def $rdi
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-BMI-NEXT: andb $15, %dil
[X86] Enable 8-bit OR with disjoint bits to convert to LEA We already support 8-bits adds in convertToThreeAddress. But we can also support 8-bit OR if the bits are disjoint. We already do this for 16/32/64. Differential Revision: https://reviews.llvm.org/D58863 llvm-svn: 355423
2019-03-06 02:37:33 +08:00
; CHECK-BMI-NEXT: andb $-16, %sil
; CHECK-BMI-NEXT: leal (%rsi,%rdi), %eax
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-BMI-NEXT: # kill: def $al killed $al killed $eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-BMI-NEXT: retq
%mx = and i8 %x, 15
%my = and i8 %y, -16
%r = or i8 %mx, %my
ret i8 %r
}
define i16 @out16_constmask(i16 %x, i16 %y) {
; CHECK-NOBMI-LABEL: out16_constmask:
; CHECK-NOBMI: # %bb.0:
; CHECK-NOBMI-NEXT: # kill: def $esi killed $esi def $rsi
; CHECK-NOBMI-NEXT: # kill: def $edi killed $edi def $rdi
; CHECK-NOBMI-NEXT: andl $3855, %edi # imm = 0xF0F
; CHECK-NOBMI-NEXT: andl $-3856, %esi # imm = 0xF0F0
; CHECK-NOBMI-NEXT: leal (%rsi,%rdi), %eax
; CHECK-NOBMI-NEXT: # kill: def $ax killed $ax killed $eax
; CHECK-NOBMI-NEXT: retq
;
; CHECK-BMI-LABEL: out16_constmask:
; CHECK-BMI: # %bb.0:
; CHECK-BMI-NEXT: # kill: def $esi killed $esi def $rsi
; CHECK-BMI-NEXT: # kill: def $edi killed $edi def $rdi
; CHECK-BMI-NEXT: andl $3855, %edi # imm = 0xF0F
; CHECK-BMI-NEXT: andl $-3856, %esi # imm = 0xF0F0
; CHECK-BMI-NEXT: leal (%rsi,%rdi), %eax
; CHECK-BMI-NEXT: # kill: def $ax killed $ax killed $eax
; CHECK-BMI-NEXT: retq
%mx = and i16 %x, 3855
%my = and i16 %y, -3856
%r = or i16 %mx, %my
ret i16 %r
}
define i32 @out32_constmask(i32 %x, i32 %y) {
; CHECK-NOBMI-LABEL: out32_constmask:
; CHECK-NOBMI: # %bb.0:
; CHECK-NOBMI-NEXT: # kill: def $esi killed $esi def $rsi
; CHECK-NOBMI-NEXT: # kill: def $edi killed $edi def $rdi
; CHECK-NOBMI-NEXT: andl $252645135, %edi # imm = 0xF0F0F0F
; CHECK-NOBMI-NEXT: andl $-252645136, %esi # imm = 0xF0F0F0F0
; CHECK-NOBMI-NEXT: leal (%rsi,%rdi), %eax
; CHECK-NOBMI-NEXT: retq
;
; CHECK-BMI-LABEL: out32_constmask:
; CHECK-BMI: # %bb.0:
; CHECK-BMI-NEXT: # kill: def $esi killed $esi def $rsi
; CHECK-BMI-NEXT: # kill: def $edi killed $edi def $rdi
; CHECK-BMI-NEXT: andl $252645135, %edi # imm = 0xF0F0F0F
; CHECK-BMI-NEXT: andl $-252645136, %esi # imm = 0xF0F0F0F0
; CHECK-BMI-NEXT: leal (%rsi,%rdi), %eax
; CHECK-BMI-NEXT: retq
%mx = and i32 %x, 252645135
%my = and i32 %y, -252645136
%r = or i32 %mx, %my
ret i32 %r
}
define i64 @out64_constmask(i64 %x, i64 %y) {
; CHECK-NOBMI-LABEL: out64_constmask:
; CHECK-NOBMI: # %bb.0:
; CHECK-NOBMI-NEXT: movabsq $1085102592571150095, %rcx # imm = 0xF0F0F0F0F0F0F0F
; CHECK-NOBMI-NEXT: andq %rdi, %rcx
; CHECK-NOBMI-NEXT: movabsq $-1085102592571150096, %rax # imm = 0xF0F0F0F0F0F0F0F0
; CHECK-NOBMI-NEXT: andq %rsi, %rax
; CHECK-NOBMI-NEXT: orq %rcx, %rax
; CHECK-NOBMI-NEXT: retq
;
; CHECK-BMI-LABEL: out64_constmask:
; CHECK-BMI: # %bb.0:
; CHECK-BMI-NEXT: movabsq $1085102592571150095, %rcx # imm = 0xF0F0F0F0F0F0F0F
; CHECK-BMI-NEXT: andq %rdi, %rcx
; CHECK-BMI-NEXT: movabsq $-1085102592571150096, %rax # imm = 0xF0F0F0F0F0F0F0F0
; CHECK-BMI-NEXT: andq %rsi, %rax
; CHECK-BMI-NEXT: orq %rcx, %rax
; CHECK-BMI-NEXT: retq
%mx = and i64 %x, 1085102592571150095
%my = and i64 %y, -1085102592571150096
%r = or i64 %mx, %my
ret i64 %r
}
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
; Should be the same as the previous one.
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
define i8 @in8_constmask(i8 %x, i8 %y) {
; CHECK-NOBMI-LABEL: in8_constmask:
; CHECK-NOBMI: # %bb.0:
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-NOBMI-NEXT: movl %esi, %eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-NOBMI-NEXT: xorl %esi, %edi
; CHECK-NOBMI-NEXT: andb $15, %dil
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-NOBMI-NEXT: xorb %dil, %al
; CHECK-NOBMI-NEXT: # kill: def $al killed $al killed $eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-NOBMI-NEXT: retq
;
; CHECK-BMI-LABEL: in8_constmask:
; CHECK-BMI: # %bb.0:
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-BMI-NEXT: movl %esi, %eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-BMI-NEXT: xorl %esi, %edi
; CHECK-BMI-NEXT: andb $15, %dil
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-BMI-NEXT: xorb %dil, %al
; CHECK-BMI-NEXT: # kill: def $al killed $al killed $eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-BMI-NEXT: retq
%n0 = xor i8 %x, %y
%n1 = and i8 %n0, 15
%r = xor i8 %n1, %y
ret i8 %r
}
define i16 @in16_constmask(i16 %x, i16 %y) {
; CHECK-NOBMI-LABEL: in16_constmask:
; CHECK-NOBMI: # %bb.0:
; CHECK-NOBMI-NEXT: movl %edi, %eax
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-NOBMI-NEXT: xorl %esi, %eax
; CHECK-NOBMI-NEXT: andl $3855, %eax # imm = 0xF0F
; CHECK-NOBMI-NEXT: xorl %esi, %eax
; CHECK-NOBMI-NEXT: # kill: def $ax killed $ax killed $eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-NOBMI-NEXT: retq
;
; CHECK-BMI-LABEL: in16_constmask:
; CHECK-BMI: # %bb.0:
; CHECK-BMI-NEXT: movl %edi, %eax
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-BMI-NEXT: xorl %esi, %eax
; CHECK-BMI-NEXT: andl $3855, %eax # imm = 0xF0F
; CHECK-BMI-NEXT: xorl %esi, %eax
; CHECK-BMI-NEXT: # kill: def $ax killed $ax killed $eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-BMI-NEXT: retq
%n0 = xor i16 %x, %y
%n1 = and i16 %n0, 3855
%r = xor i16 %n1, %y
ret i16 %r
}
define i32 @in32_constmask(i32 %x, i32 %y) {
; CHECK-NOBMI-LABEL: in32_constmask:
; CHECK-NOBMI: # %bb.0:
; CHECK-NOBMI-NEXT: movl %edi, %eax
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-NOBMI-NEXT: xorl %esi, %eax
; CHECK-NOBMI-NEXT: andl $252645135, %eax # imm = 0xF0F0F0F
; CHECK-NOBMI-NEXT: xorl %esi, %eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-NOBMI-NEXT: retq
;
; CHECK-BMI-LABEL: in32_constmask:
; CHECK-BMI: # %bb.0:
; CHECK-BMI-NEXT: movl %edi, %eax
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-BMI-NEXT: xorl %esi, %eax
; CHECK-BMI-NEXT: andl $252645135, %eax # imm = 0xF0F0F0F
; CHECK-BMI-NEXT: xorl %esi, %eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-BMI-NEXT: retq
%n0 = xor i32 %x, %y
%n1 = and i32 %n0, 252645135
%r = xor i32 %n1, %y
ret i32 %r
}
define i64 @in64_constmask(i64 %x, i64 %y) {
; CHECK-NOBMI-LABEL: in64_constmask:
; CHECK-NOBMI: # %bb.0:
; CHECK-NOBMI-NEXT: xorq %rsi, %rdi
; CHECK-NOBMI-NEXT: movabsq $1085102592571150095, %rax # imm = 0xF0F0F0F0F0F0F0F
; CHECK-NOBMI-NEXT: andq %rdi, %rax
; CHECK-NOBMI-NEXT: xorq %rsi, %rax
; CHECK-NOBMI-NEXT: retq
;
; CHECK-BMI-LABEL: in64_constmask:
; CHECK-BMI: # %bb.0:
; CHECK-BMI-NEXT: xorq %rsi, %rdi
; CHECK-BMI-NEXT: movabsq $1085102592571150095, %rax # imm = 0xF0F0F0F0F0F0F0F
; CHECK-BMI-NEXT: andq %rdi, %rax
; CHECK-BMI-NEXT: xorq %rsi, %rax
; CHECK-BMI-NEXT: retq
%n0 = xor i64 %x, %y
%n1 = and i64 %n0, 1085102592571150095
%r = xor i64 %n1, %y
ret i64 %r
}
; ============================================================================ ;
; Constant Commutativity tests.
; ============================================================================ ;
define i32 @in_constmask_commutativity_0_1(i32 %x, i32 %y) {
; CHECK-NOBMI-LABEL: in_constmask_commutativity_0_1:
; CHECK-NOBMI: # %bb.0:
; CHECK-NOBMI-NEXT: movl %edi, %eax
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-NOBMI-NEXT: xorl %esi, %eax
; CHECK-NOBMI-NEXT: andl $252645135, %eax # imm = 0xF0F0F0F
; CHECK-NOBMI-NEXT: xorl %esi, %eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-NOBMI-NEXT: retq
;
; CHECK-BMI-LABEL: in_constmask_commutativity_0_1:
; CHECK-BMI: # %bb.0:
; CHECK-BMI-NEXT: movl %edi, %eax
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-BMI-NEXT: xorl %esi, %eax
; CHECK-BMI-NEXT: andl $252645135, %eax # imm = 0xF0F0F0F
; CHECK-BMI-NEXT: xorl %esi, %eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-BMI-NEXT: retq
%n0 = xor i32 %x, %y
%n1 = and i32 %n0, 252645135
%r = xor i32 %y, %n1 ; swapped
ret i32 %r
}
define i32 @in_constmask_commutativity_1_0(i32 %x, i32 %y) {
; CHECK-NOBMI-LABEL: in_constmask_commutativity_1_0:
; CHECK-NOBMI: # %bb.0:
; CHECK-NOBMI-NEXT: movl %esi, %eax
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-NOBMI-NEXT: xorl %edi, %eax
; CHECK-NOBMI-NEXT: andl $252645135, %eax # imm = 0xF0F0F0F
; CHECK-NOBMI-NEXT: xorl %edi, %eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-NOBMI-NEXT: retq
;
; CHECK-BMI-LABEL: in_constmask_commutativity_1_0:
; CHECK-BMI: # %bb.0:
; CHECK-BMI-NEXT: movl %esi, %eax
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-BMI-NEXT: xorl %edi, %eax
; CHECK-BMI-NEXT: andl $252645135, %eax # imm = 0xF0F0F0F
; CHECK-BMI-NEXT: xorl %edi, %eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-BMI-NEXT: retq
%n0 = xor i32 %x, %y
%n1 = and i32 %n0, 252645135
%r = xor i32 %n1, %x ; %x instead of %y
ret i32 %r
}
define i32 @in_constmask_commutativity_1_1(i32 %x, i32 %y) {
; CHECK-NOBMI-LABEL: in_constmask_commutativity_1_1:
; CHECK-NOBMI: # %bb.0:
; CHECK-NOBMI-NEXT: movl %esi, %eax
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-NOBMI-NEXT: xorl %edi, %eax
; CHECK-NOBMI-NEXT: andl $252645135, %eax # imm = 0xF0F0F0F
; CHECK-NOBMI-NEXT: xorl %edi, %eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-NOBMI-NEXT: retq
;
; CHECK-BMI-LABEL: in_constmask_commutativity_1_1:
; CHECK-BMI: # %bb.0:
; CHECK-BMI-NEXT: movl %esi, %eax
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-BMI-NEXT: xorl %edi, %eax
; CHECK-BMI-NEXT: andl $252645135, %eax # imm = 0xF0F0F0F
; CHECK-BMI-NEXT: xorl %edi, %eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-BMI-NEXT: retq
%n0 = xor i32 %x, %y
%n1 = and i32 %n0, 252645135
%r = xor i32 %x, %n1 ; swapped, %x instead of %y
ret i32 %r
}
; ============================================================================ ;
; Y is an 'and' too.
; ============================================================================ ;
define i32 @in_complex_y0_constmask(i32 %x, i32 %y_hi, i32 %y_low) {
; CHECK-NOBMI-LABEL: in_complex_y0_constmask:
; CHECK-NOBMI: # %bb.0:
; CHECK-NOBMI-NEXT: movl %edi, %eax
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-NOBMI-NEXT: andl %edx, %esi
; CHECK-NOBMI-NEXT: xorl %esi, %eax
; CHECK-NOBMI-NEXT: andl $252645135, %eax # imm = 0xF0F0F0F
; CHECK-NOBMI-NEXT: xorl %esi, %eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-NOBMI-NEXT: retq
;
; CHECK-BMI-LABEL: in_complex_y0_constmask:
; CHECK-BMI: # %bb.0:
; CHECK-BMI-NEXT: movl %edi, %eax
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-BMI-NEXT: andl %edx, %esi
; CHECK-BMI-NEXT: xorl %esi, %eax
; CHECK-BMI-NEXT: andl $252645135, %eax # imm = 0xF0F0F0F
; CHECK-BMI-NEXT: xorl %esi, %eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-BMI-NEXT: retq
%y = and i32 %y_hi, %y_low
%n0 = xor i32 %x, %y
%n1 = and i32 %n0, 252645135
%r = xor i32 %n1, %y
ret i32 %r
}
define i32 @in_complex_y1_constmask(i32 %x, i32 %y_hi, i32 %y_low) {
; CHECK-NOBMI-LABEL: in_complex_y1_constmask:
; CHECK-NOBMI: # %bb.0:
; CHECK-NOBMI-NEXT: movl %edi, %eax
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-NOBMI-NEXT: andl %edx, %esi
; CHECK-NOBMI-NEXT: xorl %esi, %eax
; CHECK-NOBMI-NEXT: andl $252645135, %eax # imm = 0xF0F0F0F
; CHECK-NOBMI-NEXT: xorl %esi, %eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-NOBMI-NEXT: retq
;
; CHECK-BMI-LABEL: in_complex_y1_constmask:
; CHECK-BMI: # %bb.0:
; CHECK-BMI-NEXT: movl %edi, %eax
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-BMI-NEXT: andl %edx, %esi
; CHECK-BMI-NEXT: xorl %esi, %eax
; CHECK-BMI-NEXT: andl $252645135, %eax # imm = 0xF0F0F0F
; CHECK-BMI-NEXT: xorl %esi, %eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-BMI-NEXT: retq
%y = and i32 %y_hi, %y_low
%n0 = xor i32 %x, %y
%n1 = and i32 %n0, 252645135
%r = xor i32 %y, %n1
ret i32 %r
}
; ============================================================================ ;
; Negative tests. Should not be folded.
; ============================================================================ ;
; Multi-use tests.
declare void @use32(i32) nounwind
define i32 @in_multiuse_A_constmask(i32 %x, i32 %y, i32 %z) nounwind {
; CHECK-NOBMI-LABEL: in_multiuse_A_constmask:
; CHECK-NOBMI: # %bb.0:
; CHECK-NOBMI-NEXT: pushq %rbp
; CHECK-NOBMI-NEXT: pushq %rbx
; CHECK-NOBMI-NEXT: pushq %rax
; CHECK-NOBMI-NEXT: movl %esi, %ebx
; CHECK-NOBMI-NEXT: movl %edi, %ebp
; CHECK-NOBMI-NEXT: xorl %esi, %ebp
; CHECK-NOBMI-NEXT: andl $252645135, %ebp # imm = 0xF0F0F0F
; CHECK-NOBMI-NEXT: movl %ebp, %edi
; CHECK-NOBMI-NEXT: callq use32
; CHECK-NOBMI-NEXT: xorl %ebx, %ebp
; CHECK-NOBMI-NEXT: movl %ebp, %eax
; CHECK-NOBMI-NEXT: addq $8, %rsp
; CHECK-NOBMI-NEXT: popq %rbx
; CHECK-NOBMI-NEXT: popq %rbp
; CHECK-NOBMI-NEXT: retq
;
; CHECK-BMI-LABEL: in_multiuse_A_constmask:
; CHECK-BMI: # %bb.0:
; CHECK-BMI-NEXT: pushq %rbp
; CHECK-BMI-NEXT: pushq %rbx
; CHECK-BMI-NEXT: pushq %rax
; CHECK-BMI-NEXT: movl %esi, %ebx
; CHECK-BMI-NEXT: movl %edi, %ebp
; CHECK-BMI-NEXT: xorl %esi, %ebp
; CHECK-BMI-NEXT: andl $252645135, %ebp # imm = 0xF0F0F0F
; CHECK-BMI-NEXT: movl %ebp, %edi
; CHECK-BMI-NEXT: callq use32
; CHECK-BMI-NEXT: xorl %ebx, %ebp
; CHECK-BMI-NEXT: movl %ebp, %eax
; CHECK-BMI-NEXT: addq $8, %rsp
; CHECK-BMI-NEXT: popq %rbx
; CHECK-BMI-NEXT: popq %rbp
; CHECK-BMI-NEXT: retq
%n0 = xor i32 %x, %y
%n1 = and i32 %n0, 252645135
call void @use32(i32 %n1)
%r = xor i32 %n1, %y
ret i32 %r
}
define i32 @in_multiuse_B_constmask(i32 %x, i32 %y, i32 %z) nounwind {
; CHECK-NOBMI-LABEL: in_multiuse_B_constmask:
; CHECK-NOBMI: # %bb.0:
; CHECK-NOBMI-NEXT: pushq %rbp
; CHECK-NOBMI-NEXT: pushq %rbx
; CHECK-NOBMI-NEXT: pushq %rax
; CHECK-NOBMI-NEXT: movl %esi, %ebx
; CHECK-NOBMI-NEXT: xorl %esi, %edi
; CHECK-NOBMI-NEXT: movl %edi, %ebp
; CHECK-NOBMI-NEXT: andl $252645135, %ebp # imm = 0xF0F0F0F
; CHECK-NOBMI-NEXT: callq use32
; CHECK-NOBMI-NEXT: xorl %ebx, %ebp
; CHECK-NOBMI-NEXT: movl %ebp, %eax
; CHECK-NOBMI-NEXT: addq $8, %rsp
; CHECK-NOBMI-NEXT: popq %rbx
; CHECK-NOBMI-NEXT: popq %rbp
; CHECK-NOBMI-NEXT: retq
;
; CHECK-BMI-LABEL: in_multiuse_B_constmask:
; CHECK-BMI: # %bb.0:
; CHECK-BMI-NEXT: pushq %rbp
; CHECK-BMI-NEXT: pushq %rbx
; CHECK-BMI-NEXT: pushq %rax
; CHECK-BMI-NEXT: movl %esi, %ebx
; CHECK-BMI-NEXT: xorl %esi, %edi
; CHECK-BMI-NEXT: movl %edi, %ebp
; CHECK-BMI-NEXT: andl $252645135, %ebp # imm = 0xF0F0F0F
; CHECK-BMI-NEXT: callq use32
; CHECK-BMI-NEXT: xorl %ebx, %ebp
; CHECK-BMI-NEXT: movl %ebp, %eax
; CHECK-BMI-NEXT: addq $8, %rsp
; CHECK-BMI-NEXT: popq %rbx
; CHECK-BMI-NEXT: popq %rbp
; CHECK-BMI-NEXT: retq
%n0 = xor i32 %x, %y
%n1 = and i32 %n0, 252645135
call void @use32(i32 %n0)
%r = xor i32 %n1, %y
ret i32 %r
}
; Various bad variants
define i32 @n0_badconstmask(i32 %x, i32 %y) {
; CHECK-NOBMI-LABEL: n0_badconstmask:
; CHECK-NOBMI: # %bb.0:
; CHECK-NOBMI-NEXT: movl %esi, %eax
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-NOBMI-NEXT: andl $252645135, %edi # imm = 0xF0F0F0F
; CHECK-NOBMI-NEXT: andl $-252645135, %eax # imm = 0xF0F0F0F1
; CHECK-NOBMI-NEXT: orl %edi, %eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-NOBMI-NEXT: retq
;
; CHECK-BMI-LABEL: n0_badconstmask:
; CHECK-BMI: # %bb.0:
; CHECK-BMI-NEXT: movl %esi, %eax
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-BMI-NEXT: andl $252645135, %edi # imm = 0xF0F0F0F
; CHECK-BMI-NEXT: andl $-252645135, %eax # imm = 0xF0F0F0F1
; CHECK-BMI-NEXT: orl %edi, %eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-BMI-NEXT: retq
%mx = and i32 %x, 252645135
%my = and i32 %y, -252645135 ; instead of -252645136
%r = or i32 %mx, %my
ret i32 %r
}
define i32 @n1_thirdvar_constmask(i32 %x, i32 %y, i32 %z) {
; CHECK-NOBMI-LABEL: n1_thirdvar_constmask:
; CHECK-NOBMI: # %bb.0:
; CHECK-NOBMI-NEXT: movl %edi, %eax
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-NOBMI-NEXT: xorl %esi, %eax
; CHECK-NOBMI-NEXT: andl $252645135, %eax # imm = 0xF0F0F0F
; CHECK-NOBMI-NEXT: xorl %edx, %eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-NOBMI-NEXT: retq
;
; CHECK-BMI-LABEL: n1_thirdvar_constmask:
; CHECK-BMI: # %bb.0:
; CHECK-BMI-NEXT: movl %edi, %eax
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-BMI-NEXT: xorl %esi, %eax
; CHECK-BMI-NEXT: andl $252645135, %eax # imm = 0xF0F0F0F
; CHECK-BMI-NEXT: xorl %edx, %eax
[X86][AArch64][NFC] Add tests for masked merge unfolding Summary: This is [[ https://bugs.llvm.org/show_bug.cgi?id=37104 | PR37104 ]]. [[ https://bugs.llvm.org/show_bug.cgi?id=6773 | PR6773 ]] will introduce an IR canonicalization that is likely bad for the end assembly. Previously, `andl`+`andn`/`andps`+`andnps` / `bic`/`bsl` would be generated. (see `@out`) Now, they would no longer be generated (see `@in`). I'm guessing `llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp` should be able to unfold this. Reviewers: spatel, craig.topper, RKSimon, javed.absar Reviewed By: spatel Subscribers: nemanjai, rengolin, javed.absar, kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D45563 llvm-svn: 330645
2018-04-24 04:38:42 +08:00
; CHECK-BMI-NEXT: retq
%n0 = xor i32 %x, %y
%n1 = and i32 %n0, 252645135
%r = xor i32 %n1, %z ; instead of %y
ret i32 %r
}