RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-23 11:13:28 +01:00
Code Issues Projects Releases Wiki Activity
37342b5ed3
llvm-mirror/test/CodeGen/RISCV/select-optimize-multiple.mir

281 lines
11 KiB
Plaintext
Raw Normal View History

[RISCV] Optimize emission of SELECT sequences This patch optimizes the emission of a sequence of SELECTs with the same condition, avoiding the insertion of unnecessary control flow. Such a sequence often occurs when a SELECT of values wider than XLEN is legalized into two SELECTs with legal types. We have identified several use cases where the SELECTs could be interleaved with other instructions. Therefore, we extend the sequence to include non-SELECT instructions if we are able to detect that the non-SELECT instructions do not impact the optimization. This patch supersedes https://reviews.llvm.org/D59096, which attempted to address this issue by introducing a new SelectionDAG node. Hat tip to Eli Friedman for his feedback on how to best handle this issue. Differential Revision: https://reviews.llvm.org/D59355 Patch by Luís Marques. llvm-svn: 356741
2019-03-22 11:45:03 +01:00
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
[RISCV] Fix test after r363757 r363757 renamed ExpandISelPseudo to FinalizeISel, so the RUN line in select-optimize-multiple.mir needed updating to refer to finalize-isel. llvm-svn: 363762
2019-06-19 05:18:48 +02:00
# RUN: llc -mtriple=riscv32 -run-pass=finalize-isel -simplify-mir -o - %s \
[RISCV] Optimize emission of SELECT sequences This patch optimizes the emission of a sequence of SELECTs with the same condition, avoiding the insertion of unnecessary control flow. Such a sequence often occurs when a SELECT of values wider than XLEN is legalized into two SELECTs with legal types. We have identified several use cases where the SELECTs could be interleaved with other instructions. Therefore, we extend the sequence to include non-SELECT instructions if we are able to detect that the non-SELECT instructions do not impact the optimization. This patch supersedes https://reviews.llvm.org/D59096, which attempted to address this issue by introducing a new SelectionDAG node. Hat tip to Eli Friedman for his feedback on how to best handle this issue. Differential Revision: https://reviews.llvm.org/D59355 Patch by Luís Marques. llvm-svn: 356741
2019-03-22 11:45:03 +01:00
# RUN: | FileCheck -check-prefix=RV32I %s
[RISCV][NFC] Increase test coverage of Zbt extension Add Zbt (ternary) extension code generation to the select lowering tests since it can have a significant impact on how select is lowered. While we are here make the neg-abs commands more consistent with the other tests. Reviewed By: lenary Differential Revision: https://reviews.llvm.org/D94798
2021-01-18 18:30:19 +01:00
# RUN: llc -mtriple=riscv32 -mattr=+experimental-zbt -run-pass=finalize-isel -simplify-mir -o - %s \
# RUN: | FileCheck -check-prefix=RV32IBT %s
[RISCV] Fix test after r363757 r363757 renamed ExpandISelPseudo to FinalizeISel, so the RUN line in select-optimize-multiple.mir needed updating to refer to finalize-isel. llvm-svn: 363762
2019-06-19 05:18:48 +02:00
# RUN: llc -mtriple=riscv64 -run-pass=finalize-isel -simplify-mir -o - %s \
[RISCV] Optimize emission of SELECT sequences This patch optimizes the emission of a sequence of SELECTs with the same condition, avoiding the insertion of unnecessary control flow. Such a sequence often occurs when a SELECT of values wider than XLEN is legalized into two SELECTs with legal types. We have identified several use cases where the SELECTs could be interleaved with other instructions. Therefore, we extend the sequence to include non-SELECT instructions if we are able to detect that the non-SELECT instructions do not impact the optimization. This patch supersedes https://reviews.llvm.org/D59096, which attempted to address this issue by introducing a new SelectionDAG node. Hat tip to Eli Friedman for his feedback on how to best handle this issue. Differential Revision: https://reviews.llvm.org/D59355 Patch by Luís Marques. llvm-svn: 356741
2019-03-22 11:45:03 +01:00
# RUN: | FileCheck -check-prefix=RV64I %s
[RISCV][NFC] Increase test coverage of Zbt extension Add Zbt (ternary) extension code generation to the select lowering tests since it can have a significant impact on how select is lowered. While we are here make the neg-abs commands more consistent with the other tests. Reviewed By: lenary Differential Revision: https://reviews.llvm.org/D94798
2021-01-18 18:30:19 +01:00
# RUN: llc -mtriple=riscv64 -mattr=+experimental-zbt -run-pass=finalize-isel -simplify-mir -o - %s \
# RUN: | FileCheck -check-prefix=RV64IBT %s
[RISCV] Optimize emission of SELECT sequences This patch optimizes the emission of a sequence of SELECTs with the same condition, avoiding the insertion of unnecessary control flow. Such a sequence often occurs when a SELECT of values wider than XLEN is legalized into two SELECTs with legal types. We have identified several use cases where the SELECTs could be interleaved with other instructions. Therefore, we extend the sequence to include non-SELECT instructions if we are able to detect that the non-SELECT instructions do not impact the optimization. This patch supersedes https://reviews.llvm.org/D59096, which attempted to address this issue by introducing a new SelectionDAG node. Hat tip to Eli Friedman for his feedback on how to best handle this issue. Differential Revision: https://reviews.llvm.org/D59355 Patch by Luís Marques. llvm-svn: 356741
2019-03-22 11:45:03 +01:00
# Provide dummy definitions of functions and just enough metadata to create a
# DBG_VALUE.
--- |
define void @cmov_interleaved_bad() {
ret void
}
define void @cmov_interleaved_debug_value() {
ret void
}
...
---
# Here we have a sequence of select instructions with a non-select instruction
# in the middle. Because the non-select depends on the result of a previous
# select, we cannot optimize the sequence to share control-flow.
name: cmov_interleaved_bad
[Alignment] Use llvm::Align in MachineFunction and TargetLowering - fixes mir parsing Summary: This catches malformed mir files which specify alignment as log2 instead of pow2. See https://reviews.llvm.org/D65945 for reference, This is patch is part of a series to introduce an Alignment type. See this thread for context: http://lists.llvm.org/pipermail/llvm-dev/2019-July/133851.html See this patch for the introduction of the type: https://reviews.llvm.org/D64790 Reviewers: courbet Subscribers: MatzeB, qcolombet, dschuff, arsenm, sdardis, nemanjai, jvesely, nhaehnle, hiraditya, kbarton, asb, rbar, johnrusso, simoncook, apazos, sabuasal, niosHD, jrtc27, MaskRay, zzheng, edward-jones, atanasyan, rogfer01, MartinMosbeck, brucehoult, the_o, PkmX, jocewei, jsji, Petar.Avramovic, asbirlea, s.egerton, pzheng, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D67433 llvm-svn: 371608
2019-09-11 13:16:48 +02:00
alignment: 4
[RISCV] Optimize emission of SELECT sequences This patch optimizes the emission of a sequence of SELECTs with the same condition, avoiding the insertion of unnecessary control flow. Such a sequence often occurs when a SELECT of values wider than XLEN is legalized into two SELECTs with legal types. We have identified several use cases where the SELECTs could be interleaved with other instructions. Therefore, we extend the sequence to include non-SELECT instructions if we are able to detect that the non-SELECT instructions do not impact the optimization. This patch supersedes https://reviews.llvm.org/D59096, which attempted to address this issue by introducing a new SelectionDAG node. Hat tip to Eli Friedman for his feedback on how to best handle this issue. Differential Revision: https://reviews.llvm.org/D59355 Patch by Luís Marques. llvm-svn: 356741
2019-03-22 11:45:03 +01:00
tracksRegLiveness: true
registers:
- { id: 0, class: gpr }
- { id: 1, class: gpr }
- { id: 2, class: gpr }
- { id: 3, class: gpr }
- { id: 4, class: gpr }
- { id: 5, class: gpr }
- { id: 6, class: gpr }
- { id: 7, class: gpr }
- { id: 8, class: gpr }
- { id: 9, class: gpr }
- { id: 10, class: gpr }
liveins:
- { reg: '$x10', virtual-reg: '%0' }
- { reg: '$x11', virtual-reg: '%1' }
- { reg: '$x12', virtual-reg: '%2' }
- { reg: '$x13', virtual-reg: '%3' }
body: |
bb.0:
liveins: $x10, $x11, $x12, $x13
; RV32I-LABEL: name: cmov_interleaved_bad
; RV32I: successors: %bb.1, %bb.2
; RV32I: liveins: $x10, $x11, $x12, $x13
; RV32I: [[COPY:%[0-9]+]]:gpr = COPY $x13
; RV32I: [[COPY1:%[0-9]+]]:gpr = COPY $x12
; RV32I: [[COPY2:%[0-9]+]]:gpr = COPY $x11
; RV32I: [[COPY3:%[0-9]+]]:gpr = COPY $x10
; RV32I: [[ANDI:%[0-9]+]]:gpr = ANDI [[COPY3]], 1
; RV32I: [[COPY4:%[0-9]+]]:gpr = COPY $x0
; RV32I: BNE [[ANDI]], [[COPY4]], %bb.2
; RV32I: .1:
; RV32I: .2:
; RV32I: successors: %bb.3, %bb.4
; RV32I: [[PHI:%[0-9]+]]:gpr = PHI [[COPY2]], %bb.0, [[COPY1]], %bb.1
; RV32I: [[ADDI:%[0-9]+]]:gpr = ADDI [[PHI]], 1
; RV32I: BNE [[ANDI]], [[COPY4]], %bb.4
; RV32I: .3:
; RV32I: .4:
; RV32I: [[PHI1:%[0-9]+]]:gpr = PHI [[COPY]], %bb.2, [[COPY1]], %bb.3
; RV32I: [[ADD:%[0-9]+]]:gpr = ADD [[PHI]], killed [[PHI1]]
; RV32I: $x10 = COPY [[ADD]]
; RV32I: PseudoRET implicit $x10
[RISCV][NFC] Increase test coverage of Zbt extension Add Zbt (ternary) extension code generation to the select lowering tests since it can have a significant impact on how select is lowered. While we are here make the neg-abs commands more consistent with the other tests. Reviewed By: lenary Differential Revision: https://reviews.llvm.org/D94798
2021-01-18 18:30:19 +01:00
; RV32IBT-LABEL: name: cmov_interleaved_bad
; RV32IBT: successors: %bb.1, %bb.2
; RV32IBT: liveins: $x10, $x11, $x12, $x13
; RV32IBT: [[COPY:%[0-9]+]]:gpr = COPY $x13
; RV32IBT: [[COPY1:%[0-9]+]]:gpr = COPY $x12
; RV32IBT: [[COPY2:%[0-9]+]]:gpr = COPY $x11
; RV32IBT: [[COPY3:%[0-9]+]]:gpr = COPY $x10
; RV32IBT: [[ANDI:%[0-9]+]]:gpr = ANDI [[COPY3]], 1
; RV32IBT: [[COPY4:%[0-9]+]]:gpr = COPY $x0
; RV32IBT: BNE [[ANDI]], [[COPY4]], %bb.2
; RV32IBT: .1:
; RV32IBT: .2:
; RV32IBT: successors: %bb.3, %bb.4
; RV32IBT: [[PHI:%[0-9]+]]:gpr = PHI [[COPY2]], %bb.0, [[COPY1]], %bb.1
; RV32IBT: [[ADDI:%[0-9]+]]:gpr = ADDI [[PHI]], 1
; RV32IBT: BNE [[ANDI]], [[COPY4]], %bb.4
; RV32IBT: .3:
; RV32IBT: .4:
; RV32IBT: [[PHI1:%[0-9]+]]:gpr = PHI [[COPY]], %bb.2, [[COPY1]], %bb.3
; RV32IBT: [[ADD:%[0-9]+]]:gpr = ADD [[PHI]], killed [[PHI1]]
; RV32IBT: $x10 = COPY [[ADD]]
; RV32IBT: PseudoRET implicit $x10
[RISCV] Optimize emission of SELECT sequences This patch optimizes the emission of a sequence of SELECTs with the same condition, avoiding the insertion of unnecessary control flow. Such a sequence often occurs when a SELECT of values wider than XLEN is legalized into two SELECTs with legal types. We have identified several use cases where the SELECTs could be interleaved with other instructions. Therefore, we extend the sequence to include non-SELECT instructions if we are able to detect that the non-SELECT instructions do not impact the optimization. This patch supersedes https://reviews.llvm.org/D59096, which attempted to address this issue by introducing a new SelectionDAG node. Hat tip to Eli Friedman for his feedback on how to best handle this issue. Differential Revision: https://reviews.llvm.org/D59355 Patch by Luís Marques. llvm-svn: 356741
2019-03-22 11:45:03 +01:00
; RV64I-LABEL: name: cmov_interleaved_bad
; RV64I: successors: %bb.1, %bb.2
; RV64I: liveins: $x10, $x11, $x12, $x13
; RV64I: [[COPY:%[0-9]+]]:gpr = COPY $x13
; RV64I: [[COPY1:%[0-9]+]]:gpr = COPY $x12
; RV64I: [[COPY2:%[0-9]+]]:gpr = COPY $x11
; RV64I: [[COPY3:%[0-9]+]]:gpr = COPY $x10
; RV64I: [[ANDI:%[0-9]+]]:gpr = ANDI [[COPY3]], 1
; RV64I: [[COPY4:%[0-9]+]]:gpr = COPY $x0
; RV64I: BNE [[ANDI]], [[COPY4]], %bb.2
; RV64I: .1:
; RV64I: .2:
; RV64I: successors: %bb.3, %bb.4
; RV64I: [[PHI:%[0-9]+]]:gpr = PHI [[COPY2]], %bb.0, [[COPY1]], %bb.1
; RV64I: [[ADDI:%[0-9]+]]:gpr = ADDI [[PHI]], 1
; RV64I: BNE [[ANDI]], [[COPY4]], %bb.4
; RV64I: .3:
; RV64I: .4:
; RV64I: [[PHI1:%[0-9]+]]:gpr = PHI [[COPY]], %bb.2, [[COPY1]], %bb.3
; RV64I: [[ADD:%[0-9]+]]:gpr = ADD [[PHI]], killed [[PHI1]]
; RV64I: $x10 = COPY [[ADD]]
; RV64I: PseudoRET implicit $x10
[RISCV][NFC] Increase test coverage of Zbt extension Add Zbt (ternary) extension code generation to the select lowering tests since it can have a significant impact on how select is lowered. While we are here make the neg-abs commands more consistent with the other tests. Reviewed By: lenary Differential Revision: https://reviews.llvm.org/D94798
2021-01-18 18:30:19 +01:00
; RV64IBT-LABEL: name: cmov_interleaved_bad
; RV64IBT: successors: %bb.1, %bb.2
; RV64IBT: liveins: $x10, $x11, $x12, $x13
; RV64IBT: [[COPY:%[0-9]+]]:gpr = COPY $x13
; RV64IBT: [[COPY1:%[0-9]+]]:gpr = COPY $x12
; RV64IBT: [[COPY2:%[0-9]+]]:gpr = COPY $x11
; RV64IBT: [[COPY3:%[0-9]+]]:gpr = COPY $x10
; RV64IBT: [[ANDI:%[0-9]+]]:gpr = ANDI [[COPY3]], 1
; RV64IBT: [[COPY4:%[0-9]+]]:gpr = COPY $x0
; RV64IBT: BNE [[ANDI]], [[COPY4]], %bb.2
; RV64IBT: .1:
; RV64IBT: .2:
; RV64IBT: successors: %bb.3, %bb.4
; RV64IBT: [[PHI:%[0-9]+]]:gpr = PHI [[COPY2]], %bb.0, [[COPY1]], %bb.1
; RV64IBT: [[ADDI:%[0-9]+]]:gpr = ADDI [[PHI]], 1
; RV64IBT: BNE [[ANDI]], [[COPY4]], %bb.4
; RV64IBT: .3:
; RV64IBT: .4:
; RV64IBT: [[PHI1:%[0-9]+]]:gpr = PHI [[COPY]], %bb.2, [[COPY1]], %bb.3
; RV64IBT: [[ADD:%[0-9]+]]:gpr = ADD [[PHI]], killed [[PHI1]]
; RV64IBT: $x10 = COPY [[ADD]]
; RV64IBT: PseudoRET implicit $x10
[RISCV] Optimize emission of SELECT sequences This patch optimizes the emission of a sequence of SELECTs with the same condition, avoiding the insertion of unnecessary control flow. Such a sequence often occurs when a SELECT of values wider than XLEN is legalized into two SELECTs with legal types. We have identified several use cases where the SELECTs could be interleaved with other instructions. Therefore, we extend the sequence to include non-SELECT instructions if we are able to detect that the non-SELECT instructions do not impact the optimization. This patch supersedes https://reviews.llvm.org/D59096, which attempted to address this issue by introducing a new SelectionDAG node. Hat tip to Eli Friedman for his feedback on how to best handle this issue. Differential Revision: https://reviews.llvm.org/D59355 Patch by Luís Marques. llvm-svn: 356741
2019-03-22 11:45:03 +01:00
%3:gpr = COPY $x13
%2:gpr = COPY $x12
%1:gpr = COPY $x11
%0:gpr = COPY $x10
%5:gpr = ANDI %0, 1
%6:gpr = COPY $x0
%7:gpr = Select_GPR_Using_CC_GPR %5, %6, 22, %1, %2
%8:gpr = ADDI %7, 1
%9:gpr = Select_GPR_Using_CC_GPR %5, %6, 22, %3, %2
%10:gpr = ADD %7, killed %9
$x10 = COPY %10
PseudoRET implicit $x10
...
---
# Demonstrate that debug info associated with selects is correctly moved to
# the tail basic block, while debug info associated with non-selects is left
# in the head basic block.
name: cmov_interleaved_debug_value
[Alignment] Use llvm::Align in MachineFunction and TargetLowering - fixes mir parsing Summary: This catches malformed mir files which specify alignment as log2 instead of pow2. See https://reviews.llvm.org/D65945 for reference, This is patch is part of a series to introduce an Alignment type. See this thread for context: http://lists.llvm.org/pipermail/llvm-dev/2019-July/133851.html See this patch for the introduction of the type: https://reviews.llvm.org/D64790 Reviewers: courbet Subscribers: MatzeB, qcolombet, dschuff, arsenm, sdardis, nemanjai, jvesely, nhaehnle, hiraditya, kbarton, asb, rbar, johnrusso, simoncook, apazos, sabuasal, niosHD, jrtc27, MaskRay, zzheng, edward-jones, atanasyan, rogfer01, MartinMosbeck, brucehoult, the_o, PkmX, jocewei, jsji, Petar.Avramovic, asbirlea, s.egerton, pzheng, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D67433 llvm-svn: 371608
2019-09-11 13:16:48 +02:00
alignment: 4
[RISCV] Optimize emission of SELECT sequences This patch optimizes the emission of a sequence of SELECTs with the same condition, avoiding the insertion of unnecessary control flow. Such a sequence often occurs when a SELECT of values wider than XLEN is legalized into two SELECTs with legal types. We have identified several use cases where the SELECTs could be interleaved with other instructions. Therefore, we extend the sequence to include non-SELECT instructions if we are able to detect that the non-SELECT instructions do not impact the optimization. This patch supersedes https://reviews.llvm.org/D59096, which attempted to address this issue by introducing a new SelectionDAG node. Hat tip to Eli Friedman for his feedback on how to best handle this issue. Differential Revision: https://reviews.llvm.org/D59355 Patch by Luís Marques. llvm-svn: 356741
2019-03-22 11:45:03 +01:00
tracksRegLiveness: true
registers:
- { id: 0, class: gpr }
- { id: 1, class: gpr }
- { id: 2, class: gpr }
- { id: 3, class: gpr }
- { id: 4, class: gpr }
- { id: 5, class: gpr }
- { id: 6, class: gpr }
- { id: 7, class: gpr }
- { id: 8, class: gpr }
- { id: 9, class: gpr }
- { id: 10, class: gpr }
liveins:
- { reg: '$x10', virtual-reg: '%0' }
- { reg: '$x11', virtual-reg: '%1' }
- { reg: '$x12', virtual-reg: '%2' }
- { reg: '$x13', virtual-reg: '%3' }
body: |
bb.0:
liveins: $x10, $x11, $x12, $x13
; RV32I-LABEL: name: cmov_interleaved_debug_value
; RV32I: successors: %bb.1, %bb.2
; RV32I: liveins: $x10, $x11, $x12, $x13
; RV32I: [[COPY:%[0-9]+]]:gpr = COPY $x13
; RV32I: [[COPY1:%[0-9]+]]:gpr = COPY $x12
; RV32I: [[COPY2:%[0-9]+]]:gpr = COPY $x11
; RV32I: [[COPY3:%[0-9]+]]:gpr = COPY $x10
; RV32I: [[ANDI:%[0-9]+]]:gpr = ANDI [[COPY3]], 1
; RV32I: [[COPY4:%[0-9]+]]:gpr = COPY $x0
; RV32I: [[ADDI:%[0-9]+]]:gpr = ADDI [[COPY3]], 1
[MachineVerifier] Verify that a DBG_VALUE has a debug location Summary: Verify that each DBG_VALUE has a debug location. This is required by LiveDebugValues, and perhaps by other late passes. There's an exception for tests: lots of tests use a two-operand form of DBG_VALUE for convenience. There's no reason to prevent that. This is an extension of D80665, but there's no dependency. Reviewers: aprantl, jmorse, davide, chrisjackson Subscribers: hiraditya, asb, rbar, johnrusso, simoncook, sabuasal, niosHD, jrtc27, MaskRay, zzheng, edward-jones, rogfer01, MartinMosbeck, brucehoult, the_o, PkmX, jocewei, Jim, lenary, s.egerton, pzheng, sameer.abuasal, apazos, luismarques, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D80670
2020-05-28 00:44:10 +02:00
; RV32I: DBG_VALUE [[ADDI]], $noreg
[RISCV] Optimize emission of SELECT sequences This patch optimizes the emission of a sequence of SELECTs with the same condition, avoiding the insertion of unnecessary control flow. Such a sequence often occurs when a SELECT of values wider than XLEN is legalized into two SELECTs with legal types. We have identified several use cases where the SELECTs could be interleaved with other instructions. Therefore, we extend the sequence to include non-SELECT instructions if we are able to detect that the non-SELECT instructions do not impact the optimization. This patch supersedes https://reviews.llvm.org/D59096, which attempted to address this issue by introducing a new SelectionDAG node. Hat tip to Eli Friedman for his feedback on how to best handle this issue. Differential Revision: https://reviews.llvm.org/D59355 Patch by Luís Marques. llvm-svn: 356741
2019-03-22 11:45:03 +01:00
; RV32I: BNE [[ANDI]], [[COPY4]], %bb.2
; RV32I: .1:
; RV32I: .2:
; RV32I: [[PHI:%[0-9]+]]:gpr = PHI [[COPY2]], %bb.0, [[COPY1]], %bb.1
; RV32I: [[PHI1:%[0-9]+]]:gpr = PHI [[COPY]], %bb.0, [[COPY1]], %bb.1
[MachineVerifier] Verify that a DBG_VALUE has a debug location Summary: Verify that each DBG_VALUE has a debug location. This is required by LiveDebugValues, and perhaps by other late passes. There's an exception for tests: lots of tests use a two-operand form of DBG_VALUE for convenience. There's no reason to prevent that. This is an extension of D80665, but there's no dependency. Reviewers: aprantl, jmorse, davide, chrisjackson Subscribers: hiraditya, asb, rbar, johnrusso, simoncook, sabuasal, niosHD, jrtc27, MaskRay, zzheng, edward-jones, rogfer01, MartinMosbeck, brucehoult, the_o, PkmX, jocewei, Jim, lenary, s.egerton, pzheng, sameer.abuasal, apazos, luismarques, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D80670
2020-05-28 00:44:10 +02:00
; RV32I: DBG_VALUE [[PHI]], $noreg
; RV32I: DBG_VALUE [[PHI1]], $noreg
[RISCV] Optimize emission of SELECT sequences This patch optimizes the emission of a sequence of SELECTs with the same condition, avoiding the insertion of unnecessary control flow. Such a sequence often occurs when a SELECT of values wider than XLEN is legalized into two SELECTs with legal types. We have identified several use cases where the SELECTs could be interleaved with other instructions. Therefore, we extend the sequence to include non-SELECT instructions if we are able to detect that the non-SELECT instructions do not impact the optimization. This patch supersedes https://reviews.llvm.org/D59096, which attempted to address this issue by introducing a new SelectionDAG node. Hat tip to Eli Friedman for his feedback on how to best handle this issue. Differential Revision: https://reviews.llvm.org/D59355 Patch by Luís Marques. llvm-svn: 356741
2019-03-22 11:45:03 +01:00
; RV32I: [[ADD:%[0-9]+]]:gpr = ADD [[PHI]], killed [[PHI1]]
; RV32I: $x10 = COPY [[ADD]]
; RV32I: PseudoRET implicit $x10
[RISCV][NFC] Increase test coverage of Zbt extension Add Zbt (ternary) extension code generation to the select lowering tests since it can have a significant impact on how select is lowered. While we are here make the neg-abs commands more consistent with the other tests. Reviewed By: lenary Differential Revision: https://reviews.llvm.org/D94798
2021-01-18 18:30:19 +01:00
; RV32IBT-LABEL: name: cmov_interleaved_debug_value
; RV32IBT: successors: %bb.1, %bb.2
; RV32IBT: liveins: $x10, $x11, $x12, $x13
; RV32IBT: [[COPY:%[0-9]+]]:gpr = COPY $x13
; RV32IBT: [[COPY1:%[0-9]+]]:gpr = COPY $x12
; RV32IBT: [[COPY2:%[0-9]+]]:gpr = COPY $x11
; RV32IBT: [[COPY3:%[0-9]+]]:gpr = COPY $x10
; RV32IBT: [[ANDI:%[0-9]+]]:gpr = ANDI [[COPY3]], 1
; RV32IBT: [[COPY4:%[0-9]+]]:gpr = COPY $x0
; RV32IBT: [[ADDI:%[0-9]+]]:gpr = ADDI [[COPY3]], 1
; RV32IBT: DBG_VALUE [[ADDI]], $noreg
; RV32IBT: BNE [[ANDI]], [[COPY4]], %bb.2
; RV32IBT: .1:
; RV32IBT: .2:
; RV32IBT: [[PHI:%[0-9]+]]:gpr = PHI [[COPY2]], %bb.0, [[COPY1]], %bb.1
; RV32IBT: [[PHI1:%[0-9]+]]:gpr = PHI [[COPY]], %bb.0, [[COPY1]], %bb.1
; RV32IBT: DBG_VALUE [[PHI]], $noreg
; RV32IBT: DBG_VALUE [[PHI1]], $noreg
; RV32IBT: [[ADD:%[0-9]+]]:gpr = ADD [[PHI]], killed [[PHI1]]
; RV32IBT: $x10 = COPY [[ADD]]
; RV32IBT: PseudoRET implicit $x10
[RISCV] Optimize emission of SELECT sequences This patch optimizes the emission of a sequence of SELECTs with the same condition, avoiding the insertion of unnecessary control flow. Such a sequence often occurs when a SELECT of values wider than XLEN is legalized into two SELECTs with legal types. We have identified several use cases where the SELECTs could be interleaved with other instructions. Therefore, we extend the sequence to include non-SELECT instructions if we are able to detect that the non-SELECT instructions do not impact the optimization. This patch supersedes https://reviews.llvm.org/D59096, which attempted to address this issue by introducing a new SelectionDAG node. Hat tip to Eli Friedman for his feedback on how to best handle this issue. Differential Revision: https://reviews.llvm.org/D59355 Patch by Luís Marques. llvm-svn: 356741
2019-03-22 11:45:03 +01:00
; RV64I-LABEL: name: cmov_interleaved_debug_value
; RV64I: successors: %bb.1, %bb.2
; RV64I: liveins: $x10, $x11, $x12, $x13
; RV64I: [[COPY:%[0-9]+]]:gpr = COPY $x13
; RV64I: [[COPY1:%[0-9]+]]:gpr = COPY $x12
; RV64I: [[COPY2:%[0-9]+]]:gpr = COPY $x11
; RV64I: [[COPY3:%[0-9]+]]:gpr = COPY $x10
; RV64I: [[ANDI:%[0-9]+]]:gpr = ANDI [[COPY3]], 1
; RV64I: [[COPY4:%[0-9]+]]:gpr = COPY $x0
; RV64I: [[ADDI:%[0-9]+]]:gpr = ADDI [[COPY3]], 1
[MachineVerifier] Verify that a DBG_VALUE has a debug location Summary: Verify that each DBG_VALUE has a debug location. This is required by LiveDebugValues, and perhaps by other late passes. There's an exception for tests: lots of tests use a two-operand form of DBG_VALUE for convenience. There's no reason to prevent that. This is an extension of D80665, but there's no dependency. Reviewers: aprantl, jmorse, davide, chrisjackson Subscribers: hiraditya, asb, rbar, johnrusso, simoncook, sabuasal, niosHD, jrtc27, MaskRay, zzheng, edward-jones, rogfer01, MartinMosbeck, brucehoult, the_o, PkmX, jocewei, Jim, lenary, s.egerton, pzheng, sameer.abuasal, apazos, luismarques, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D80670
2020-05-28 00:44:10 +02:00
; RV64I: DBG_VALUE [[ADDI]], $noreg
[RISCV] Optimize emission of SELECT sequences This patch optimizes the emission of a sequence of SELECTs with the same condition, avoiding the insertion of unnecessary control flow. Such a sequence often occurs when a SELECT of values wider than XLEN is legalized into two SELECTs with legal types. We have identified several use cases where the SELECTs could be interleaved with other instructions. Therefore, we extend the sequence to include non-SELECT instructions if we are able to detect that the non-SELECT instructions do not impact the optimization. This patch supersedes https://reviews.llvm.org/D59096, which attempted to address this issue by introducing a new SelectionDAG node. Hat tip to Eli Friedman for his feedback on how to best handle this issue. Differential Revision: https://reviews.llvm.org/D59355 Patch by Luís Marques. llvm-svn: 356741
2019-03-22 11:45:03 +01:00
; RV64I: BNE [[ANDI]], [[COPY4]], %bb.2
; RV64I: .1:
; RV64I: .2:
; RV64I: [[PHI:%[0-9]+]]:gpr = PHI [[COPY2]], %bb.0, [[COPY1]], %bb.1
; RV64I: [[PHI1:%[0-9]+]]:gpr = PHI [[COPY]], %bb.0, [[COPY1]], %bb.1
[MachineVerifier] Verify that a DBG_VALUE has a debug location Summary: Verify that each DBG_VALUE has a debug location. This is required by LiveDebugValues, and perhaps by other late passes. There's an exception for tests: lots of tests use a two-operand form of DBG_VALUE for convenience. There's no reason to prevent that. This is an extension of D80665, but there's no dependency. Reviewers: aprantl, jmorse, davide, chrisjackson Subscribers: hiraditya, asb, rbar, johnrusso, simoncook, sabuasal, niosHD, jrtc27, MaskRay, zzheng, edward-jones, rogfer01, MartinMosbeck, brucehoult, the_o, PkmX, jocewei, Jim, lenary, s.egerton, pzheng, sameer.abuasal, apazos, luismarques, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D80670
2020-05-28 00:44:10 +02:00
; RV64I: DBG_VALUE [[PHI]], $noreg
; RV64I: DBG_VALUE [[PHI1]], $noreg
[RISCV] Optimize emission of SELECT sequences This patch optimizes the emission of a sequence of SELECTs with the same condition, avoiding the insertion of unnecessary control flow. Such a sequence often occurs when a SELECT of values wider than XLEN is legalized into two SELECTs with legal types. We have identified several use cases where the SELECTs could be interleaved with other instructions. Therefore, we extend the sequence to include non-SELECT instructions if we are able to detect that the non-SELECT instructions do not impact the optimization. This patch supersedes https://reviews.llvm.org/D59096, which attempted to address this issue by introducing a new SelectionDAG node. Hat tip to Eli Friedman for his feedback on how to best handle this issue. Differential Revision: https://reviews.llvm.org/D59355 Patch by Luís Marques. llvm-svn: 356741
2019-03-22 11:45:03 +01:00
; RV64I: [[ADD:%[0-9]+]]:gpr = ADD [[PHI]], killed [[PHI1]]
; RV64I: $x10 = COPY [[ADD]]
; RV64I: PseudoRET implicit $x10
[RISCV][NFC] Increase test coverage of Zbt extension Add Zbt (ternary) extension code generation to the select lowering tests since it can have a significant impact on how select is lowered. While we are here make the neg-abs commands more consistent with the other tests. Reviewed By: lenary Differential Revision: https://reviews.llvm.org/D94798
2021-01-18 18:30:19 +01:00
; RV64IBT-LABEL: name: cmov_interleaved_debug_value
; RV64IBT: successors: %bb.1, %bb.2
; RV64IBT: liveins: $x10, $x11, $x12, $x13
; RV64IBT: [[COPY:%[0-9]+]]:gpr = COPY $x13
; RV64IBT: [[COPY1:%[0-9]+]]:gpr = COPY $x12
; RV64IBT: [[COPY2:%[0-9]+]]:gpr = COPY $x11
; RV64IBT: [[COPY3:%[0-9]+]]:gpr = COPY $x10
; RV64IBT: [[ANDI:%[0-9]+]]:gpr = ANDI [[COPY3]], 1
; RV64IBT: [[COPY4:%[0-9]+]]:gpr = COPY $x0
; RV64IBT: [[ADDI:%[0-9]+]]:gpr = ADDI [[COPY3]], 1
; RV64IBT: DBG_VALUE [[ADDI]], $noreg
; RV64IBT: BNE [[ANDI]], [[COPY4]], %bb.2
; RV64IBT: .1:
; RV64IBT: .2:
; RV64IBT: [[PHI:%[0-9]+]]:gpr = PHI [[COPY2]], %bb.0, [[COPY1]], %bb.1
; RV64IBT: [[PHI1:%[0-9]+]]:gpr = PHI [[COPY]], %bb.0, [[COPY1]], %bb.1
; RV64IBT: DBG_VALUE [[PHI]], $noreg
; RV64IBT: DBG_VALUE [[PHI1]], $noreg
; RV64IBT: [[ADD:%[0-9]+]]:gpr = ADD [[PHI]], killed [[PHI1]]
; RV64IBT: $x10 = COPY [[ADD]]
; RV64IBT: PseudoRET implicit $x10
[RISCV] Optimize emission of SELECT sequences This patch optimizes the emission of a sequence of SELECTs with the same condition, avoiding the insertion of unnecessary control flow. Such a sequence often occurs when a SELECT of values wider than XLEN is legalized into two SELECTs with legal types. We have identified several use cases where the SELECTs could be interleaved with other instructions. Therefore, we extend the sequence to include non-SELECT instructions if we are able to detect that the non-SELECT instructions do not impact the optimization. This patch supersedes https://reviews.llvm.org/D59096, which attempted to address this issue by introducing a new SelectionDAG node. Hat tip to Eli Friedman for his feedback on how to best handle this issue. Differential Revision: https://reviews.llvm.org/D59355 Patch by Luís Marques. llvm-svn: 356741
2019-03-22 11:45:03 +01:00
%3:gpr = COPY $x13
%2:gpr = COPY $x12
%1:gpr = COPY $x11
%0:gpr = COPY $x10
%5:gpr = ANDI %0, 1
%6:gpr = COPY $x0
%7:gpr = Select_GPR_Using_CC_GPR %5, %6, 22, %1, %2
[MachineVerifier] Verify that a DBG_VALUE has a debug location Summary: Verify that each DBG_VALUE has a debug location. This is required by LiveDebugValues, and perhaps by other late passes. There's an exception for tests: lots of tests use a two-operand form of DBG_VALUE for convenience. There's no reason to prevent that. This is an extension of D80665, but there's no dependency. Reviewers: aprantl, jmorse, davide, chrisjackson Subscribers: hiraditya, asb, rbar, johnrusso, simoncook, sabuasal, niosHD, jrtc27, MaskRay, zzheng, edward-jones, rogfer01, MartinMosbeck, brucehoult, the_o, PkmX, jocewei, Jim, lenary, s.egerton, pzheng, sameer.abuasal, apazos, luismarques, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D80670
2020-05-28 00:44:10 +02:00
DBG_VALUE %7, $noreg
[RISCV] Optimize emission of SELECT sequences This patch optimizes the emission of a sequence of SELECTs with the same condition, avoiding the insertion of unnecessary control flow. Such a sequence often occurs when a SELECT of values wider than XLEN is legalized into two SELECTs with legal types. We have identified several use cases where the SELECTs could be interleaved with other instructions. Therefore, we extend the sequence to include non-SELECT instructions if we are able to detect that the non-SELECT instructions do not impact the optimization. This patch supersedes https://reviews.llvm.org/D59096, which attempted to address this issue by introducing a new SelectionDAG node. Hat tip to Eli Friedman for his feedback on how to best handle this issue. Differential Revision: https://reviews.llvm.org/D59355 Patch by Luís Marques. llvm-svn: 356741
2019-03-22 11:45:03 +01:00
%8:gpr = ADDI %0, 1
[MachineVerifier] Verify that a DBG_VALUE has a debug location Summary: Verify that each DBG_VALUE has a debug location. This is required by LiveDebugValues, and perhaps by other late passes. There's an exception for tests: lots of tests use a two-operand form of DBG_VALUE for convenience. There's no reason to prevent that. This is an extension of D80665, but there's no dependency. Reviewers: aprantl, jmorse, davide, chrisjackson Subscribers: hiraditya, asb, rbar, johnrusso, simoncook, sabuasal, niosHD, jrtc27, MaskRay, zzheng, edward-jones, rogfer01, MartinMosbeck, brucehoult, the_o, PkmX, jocewei, Jim, lenary, s.egerton, pzheng, sameer.abuasal, apazos, luismarques, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D80670
2020-05-28 00:44:10 +02:00
DBG_VALUE %8, $noreg
[RISCV] Optimize emission of SELECT sequences This patch optimizes the emission of a sequence of SELECTs with the same condition, avoiding the insertion of unnecessary control flow. Such a sequence often occurs when a SELECT of values wider than XLEN is legalized into two SELECTs with legal types. We have identified several use cases where the SELECTs could be interleaved with other instructions. Therefore, we extend the sequence to include non-SELECT instructions if we are able to detect that the non-SELECT instructions do not impact the optimization. This patch supersedes https://reviews.llvm.org/D59096, which attempted to address this issue by introducing a new SelectionDAG node. Hat tip to Eli Friedman for his feedback on how to best handle this issue. Differential Revision: https://reviews.llvm.org/D59355 Patch by Luís Marques. llvm-svn: 356741
2019-03-22 11:45:03 +01:00
%9:gpr = Select_GPR_Using_CC_GPR %5, %6, 22, %3, %2
[MachineVerifier] Verify that a DBG_VALUE has a debug location Summary: Verify that each DBG_VALUE has a debug location. This is required by LiveDebugValues, and perhaps by other late passes. There's an exception for tests: lots of tests use a two-operand form of DBG_VALUE for convenience. There's no reason to prevent that. This is an extension of D80665, but there's no dependency. Reviewers: aprantl, jmorse, davide, chrisjackson Subscribers: hiraditya, asb, rbar, johnrusso, simoncook, sabuasal, niosHD, jrtc27, MaskRay, zzheng, edward-jones, rogfer01, MartinMosbeck, brucehoult, the_o, PkmX, jocewei, Jim, lenary, s.egerton, pzheng, sameer.abuasal, apazos, luismarques, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D80670
2020-05-28 00:44:10 +02:00
DBG_VALUE %9, $noreg
[RISCV] Optimize emission of SELECT sequences This patch optimizes the emission of a sequence of SELECTs with the same condition, avoiding the insertion of unnecessary control flow. Such a sequence often occurs when a SELECT of values wider than XLEN is legalized into two SELECTs with legal types. We have identified several use cases where the SELECTs could be interleaved with other instructions. Therefore, we extend the sequence to include non-SELECT instructions if we are able to detect that the non-SELECT instructions do not impact the optimization. This patch supersedes https://reviews.llvm.org/D59096, which attempted to address this issue by introducing a new SelectionDAG node. Hat tip to Eli Friedman for his feedback on how to best handle this issue. Differential Revision: https://reviews.llvm.org/D59355 Patch by Luís Marques. llvm-svn: 356741
2019-03-22 11:45:03 +01:00
%10:gpr = ADD %7, killed %9
$x10 = COPY %10
PseudoRET implicit $x10
...
---
Reference in New Issue Copy Permalink