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/switch.ll

781 lines
23 KiB
LLVM
Raw Normal View History

[CodeGen] Peel off the dominant case in switch statement in lowering This patch peels off the top case in switch statement into a branch if the probability exceeds a threshold. This will help the branch prediction and avoids the extra compares when lowering into chain of branches. Differential Revision: http://reviews.llvm.org/D39262 llvm-svn: 318202
2017-11-15 05:44:09 +08:00
; RUN: llc -mtriple=x86_64-linux-gnu %s -o - -jump-table-density=40 -switch-peel-threshold=101 -verify-machineinstrs | FileCheck %s
Switch lowering: don't add incoming PHI values from skipped bit test MBB's (PR27135) After r245976, LLVM will skip the last bit test case if knows it will always be true. However, we would still erroneously update PHI nodes with incoming values from the MBB that would perform the final bit test, causing -verify-machineinstrs to fail. llvm-svn: 266479
2016-04-16 05:45:30 +08:00
; RUN: llc -mtriple=x86_64-linux-gnu %s -o - -O0 -jump-table-density=40 -verify-machineinstrs | FileCheck --check-prefix=NOOPT %s
Re-commit r235560: Switch lowering: extract jump tables and bit tests before building binary tree (PR22262) Third time's the charm. The previous commit was reverted as a reverse for-loop in SelectionDAGBuilder::lowerWorkItem did 'I--' on an iterator at the beginning of a vector, causing asserts when using debugging iterators. This commit fixes that. llvm-svn: 235608
2015-04-24 00:45:24 +08:00
declare void @g(i32)
define void @basic(i32 %x) {
entry:
switch i32 %x, label %return [
i32 3, label %bb0
i32 1, label %bb1
i32 4, label %bb1
Switch lowering: cluster adjacent fall-through cases even at -O0 It's cheap to do, and codegen is much faster if cases can be merged into clusters. llvm-svn: 236905
2015-05-09 05:23:39 +08:00
i32 5, label %bb2
Re-commit r235560: Switch lowering: extract jump tables and bit tests before building binary tree (PR22262) Third time's the charm. The previous commit was reverted as a reverse for-loop in SelectionDAGBuilder::lowerWorkItem did 'I--' on an iterator at the beginning of a vector, causing asserts when using debugging iterators. This commit fixes that. llvm-svn: 235608
2015-04-24 00:45:24 +08:00
]
bb0: tail call void @g(i32 0) br label %return
bb1: tail call void @g(i32 1) br label %return
Switch lowering: cluster adjacent fall-through cases even at -O0 It's cheap to do, and codegen is much faster if cases can be merged into clusters. llvm-svn: 236905
2015-05-09 05:23:39 +08:00
bb2: tail call void @g(i32 1) br label %return
Re-commit r235560: Switch lowering: extract jump tables and bit tests before building binary tree (PR22262) Third time's the charm. The previous commit was reverted as a reverse for-loop in SelectionDAGBuilder::lowerWorkItem did 'I--' on an iterator at the beginning of a vector, causing asserts when using debugging iterators. This commit fixes that. llvm-svn: 235608
2015-04-24 00:45:24 +08:00
return: ret void
Switch lowering: enable whole-switch jump tables at -O0. To same compile time, the analysis to find dense case-clusters in switches is not done at -O0. However, when the whole switch is dense enough, it is easy to turn it into a jump table, resulting in much faster code with no extra effort. llvm-svn: 240071
2015-06-19 06:22:30 +08:00
; Lowered as a jump table, both with and without optimization.
Re-commit r235560: Switch lowering: extract jump tables and bit tests before building binary tree (PR22262) Third time's the charm. The previous commit was reverted as a reverse for-loop in SelectionDAGBuilder::lowerWorkItem did 'I--' on an iterator at the beginning of a vector, causing asserts when using debugging iterators. This commit fixes that. llvm-svn: 235608
2015-04-24 00:45:24 +08:00
; CHECK-LABEL: basic
; CHECK: decl
; CHECK: cmpl $4
; CHECK: ja
; CHECK: jmpq *.LJTI
Switch lowering: enable whole-switch jump tables at -O0. To same compile time, the analysis to find dense case-clusters in switches is not done at -O0. However, when the whole switch is dense enough, it is easy to turn it into a jump table, resulting in much faster code with no extra effort. llvm-svn: 240071
2015-06-19 06:22:30 +08:00
; NOOPT-LABEL: basic
; NOOPT: decl
; NOOPT: subl $4
; NOOPT: ja
; NOOPT: movq .LJTI
; NOOPT: jmpq
Re-commit r235560: Switch lowering: extract jump tables and bit tests before building binary tree (PR22262) Third time's the charm. The previous commit was reverted as a reverse for-loop in SelectionDAGBuilder::lowerWorkItem did 'I--' on an iterator at the beginning of a vector, causing asserts when using debugging iterators. This commit fixes that. llvm-svn: 235608
2015-04-24 00:45:24 +08:00
}
Add support for no-jump-tables Add function soft attribute to the generation of Jump Tables in CodeGen as initial step towards clang support of gcc's no-jump-table support Reviewers: hans, echristo Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D18321 llvm-svn: 264756
2016-03-30 01:46:23 +08:00
; Should never be lowered as a jump table because of the attribute
define void @basic_nojumptable(i32 %x) "no-jump-tables"="true" {
entry:
switch i32 %x, label %return [
i32 3, label %bb0
i32 1, label %bb1
i32 4, label %bb1
i32 5, label %bb2
]
bb0: tail call void @g(i32 0) br label %return
bb1: tail call void @g(i32 1) br label %return
bb2: tail call void @g(i32 1) br label %return
return: ret void
; Lowered as a jump table, both with and without optimization.
; CHECK-LABEL: basic_nojumptable
; CHECK-NOT: jmpq *.LJTI
}
; Should be lowered as a jump table because of the attribute
define void @basic_nojumptable_false(i32 %x) "no-jump-tables"="false" {
entry:
switch i32 %x, label %return [
i32 3, label %bb0
i32 1, label %bb1
i32 4, label %bb1
i32 5, label %bb2
]
bb0: tail call void @g(i32 0) br label %return
bb1: tail call void @g(i32 1) br label %return
bb2: tail call void @g(i32 1) br label %return
return: ret void
; Lowered as a jump table, both with and without optimization.
; CHECK-LABEL: basic_nojumptable_false
; CHECK: decl
; CHECK: cmpl $4
; CHECK: ja
; CHECK: jmpq *.LJTI
}
Re-commit r235560: Switch lowering: extract jump tables and bit tests before building binary tree (PR22262) Third time's the charm. The previous commit was reverted as a reverse for-loop in SelectionDAGBuilder::lowerWorkItem did 'I--' on an iterator at the beginning of a vector, causing asserts when using debugging iterators. This commit fixes that. llvm-svn: 235608
2015-04-24 00:45:24 +08:00
define void @simple_ranges(i32 %x) {
entry:
switch i32 %x, label %return [
i32 0, label %bb0
i32 1, label %bb0
i32 2, label %bb0
i32 3, label %bb0
i32 100, label %bb1
i32 101, label %bb1
i32 102, label %bb1
i32 103, label %bb1
]
bb0: tail call void @g(i32 0) br label %return
bb1: tail call void @g(i32 1) br label %return
return: ret void
Add support for no-jump-tables Add function soft attribute to the generation of Jump Tables in CodeGen as initial step towards clang support of gcc's no-jump-table support Reviewers: hans, echristo Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D18321 llvm-svn: 264756
2016-03-30 01:46:23 +08:00
Re-commit r235560: Switch lowering: extract jump tables and bit tests before building binary tree (PR22262) Third time's the charm. The previous commit was reverted as a reverse for-loop in SelectionDAGBuilder::lowerWorkItem did 'I--' on an iterator at the beginning of a vector, causing asserts when using debugging iterators. This commit fixes that. llvm-svn: 235608
2015-04-24 00:45:24 +08:00
; Should be lowered to two range checks.
; CHECK-LABEL: simple_ranges
; CHECK: leal -100
; CHECK: cmpl $4
Distribute the weight on the edge from switch to default statement to edges generated in lowering switch. Currently, when edge weights are assigned to edges that are created when lowering switch statement, the weight on the edge to default statement (let's call it "default weight" here) is not considered. We need to distribute this weight properly. However, without value profiling, we have no idea how to distribute it. In this patch, I applied the heuristic that this weight is evenly distributed to successors. For example, given a switch statement with cases 1,2,3,5,10,11,20, and every edge from switch to each successor has weight 10. If there is a binary search tree built to test if n < 10, then its two out-edges will have weight 4x10+10/2 = 45 and 3x10 + 10/2 = 35 respectively (currently they are 40 and 30 without considering the default weight). Each distribution (which is 5 here) will be stored in each SwitchWorkListItem for further distribution. There are some exceptions: For a jump table header which doesn't have any edge to default statement, we don't distribute the default weight to it. For a bit test header which covers a contiguous range and hence has no edges to default statement, we don't distribute the default weight to it. When the branch checks a single value or a contiguous range with no edge to default statement, we don't distribute the default weight to it. In other cases, the default weight is evenly distributed to successors. Differential Revision: http://reviews.llvm.org/D12418 llvm-svn: 246522
2015-09-01 09:42:16 +08:00
; CHECK: jb
Re-commit r235560: Switch lowering: extract jump tables and bit tests before building binary tree (PR22262) Third time's the charm. The previous commit was reverted as a reverse for-loop in SelectionDAGBuilder::lowerWorkItem did 'I--' on an iterator at the beginning of a vector, causing asserts when using debugging iterators. This commit fixes that. llvm-svn: 235608
2015-04-24 00:45:24 +08:00
; CHECK: cmpl $3
; CHECK: ja
Switch lowering: cluster adjacent fall-through cases even at -O0 It's cheap to do, and codegen is much faster if cases can be merged into clusters. llvm-svn: 236905
2015-05-09 05:23:39 +08:00
; We do this even at -O0, because it's cheap and makes codegen faster.
; NOOPT-LABEL: simple_ranges
; NOOPT: subl $4
; NOOPT: jb
; NOOPT: addl $-100
; NOOPT: subl $4
; NOOPT: jb
Re-commit r235560: Switch lowering: extract jump tables and bit tests before building binary tree (PR22262) Third time's the charm. The previous commit was reverted as a reverse for-loop in SelectionDAGBuilder::lowerWorkItem did 'I--' on an iterator at the beginning of a vector, causing asserts when using debugging iterators. This commit fixes that. llvm-svn: 235608
2015-04-24 00:45:24 +08:00
}
define void @jt_is_better(i32 %x) {
entry:
switch i32 %x, label %return [
i32 0, label %bb0
i32 2, label %bb0
i32 4, label %bb0
i32 1, label %bb1
i32 3, label %bb1
i32 5, label %bb1
i32 6, label %bb2
i32 7, label %bb3
i32 8, label %bb4
]
bb0: tail call void @g(i32 0) br label %return
bb1: tail call void @g(i32 1) br label %return
bb2: tail call void @g(i32 2) br label %return
bb3: tail call void @g(i32 3) br label %return
bb4: tail call void @g(i32 4) br label %return
return: ret void
; Cases 0-5 could be lowered with two bit tests,
; but with 6-8, the whole switch is suitable for a jump table.
; CHECK-LABEL: jt_is_better
; CHECK: cmpl $8
Assign weights to edges to jump table / bit test header when lowering switch statement. Currently, when lowering switch statement and a new basic block is built for jump table / bit test header, the edge to this new block is not assigned with a correct weight. This patch collects the edge weight from all its successors and assign this sum of weights to the edge (and also the other fall-through edge). Test cases are adjusted accordingly. Differential Revision: http://reviews.llvm.org/D12166#fae6eca7 llvm-svn: 246104
2015-08-27 07:15:32 +08:00
; CHECK: ja
Re-commit r235560: Switch lowering: extract jump tables and bit tests before building binary tree (PR22262) Third time's the charm. The previous commit was reverted as a reverse for-loop in SelectionDAGBuilder::lowerWorkItem did 'I--' on an iterator at the beginning of a vector, causing asserts when using debugging iterators. This commit fixes that. llvm-svn: 235608
2015-04-24 00:45:24 +08:00
; CHECK: jmpq *.LJTI
}
define void @bt_is_better(i32 %x) {
entry:
switch i32 %x, label %return [
i32 0, label %bb0
i32 3, label %bb0
i32 6, label %bb0
i32 1, label %bb1
i32 4, label %bb1
i32 7, label %bb1
i32 2, label %bb2
i32 5, label %bb2
i32 8, label %bb2
]
bb0: tail call void @g(i32 0) br label %return
bb1: tail call void @g(i32 1) br label %return
bb2: tail call void @g(i32 2) br label %return
return: ret void
; This could be lowered as a jump table, but bit tests is more efficient.
; CHECK-LABEL: bt_is_better
Remove the final bit test during lowering switch statement if all cases in bit test cover a contiguous range. When lowering switch statement, if bit tests are used then LLVM will always generates a jump to the default statement in the last bit test. However, this is not necessary when all cases in bit tests cover a contiguous range. This is because when generating the bit tests header MBB, there is a range check that guarantees cases in bit tests won't go outside of [low, high], where low and high are minimum and maximum case values in the bit tests. This patch checks if this is the case and then doesn't emit jump to default statement and hence saves a bit test and a branch. Differential Revision: http://reviews.llvm.org/D12249 llvm-svn: 245976
2015-08-26 05:34:38 +08:00
; The bit test on 2,5,8 is unnecessary as all cases cover the rage [0, 8].
; The range check guarantees that cases other than 0,3,6 and 1,4,7 must be
; in 2,5,8.
;
Re-commit r235560: Switch lowering: extract jump tables and bit tests before building binary tree (PR22262) Third time's the charm. The previous commit was reverted as a reverse for-loop in SelectionDAGBuilder::lowerWorkItem did 'I--' on an iterator at the beginning of a vector, causing asserts when using debugging iterators. This commit fixes that. llvm-svn: 235608
2015-04-24 00:45:24 +08:00
; 73 = 2^0 + 2^3 + 2^6
; CHECK: movl $73
; CHECK: btl
; 146 = 2^1 + 2^4 + 2^7
; CHECK: movl $146
; CHECK: btl
; 292 = 2^2 + 2^5 + 2^8
Remove the final bit test during lowering switch statement if all cases in bit test cover a contiguous range. When lowering switch statement, if bit tests are used then LLVM will always generates a jump to the default statement in the last bit test. However, this is not necessary when all cases in bit tests cover a contiguous range. This is because when generating the bit tests header MBB, there is a range check that guarantees cases in bit tests won't go outside of [low, high], where low and high are minimum and maximum case values in the bit tests. This patch checks if this is the case and then doesn't emit jump to default statement and hence saves a bit test and a branch. Differential Revision: http://reviews.llvm.org/D12249 llvm-svn: 245976
2015-08-26 05:34:38 +08:00
; CHECK-NOT: movl $292
; CHECK-NOT: btl
}
define void @bt_is_better2(i32 %x) {
entry:
switch i32 %x, label %return [
i32 0, label %bb0
i32 3, label %bb0
i32 6, label %bb0
i32 1, label %bb1
i32 4, label %bb1
i32 7, label %bb1
i32 2, label %bb2
i32 8, label %bb2
]
bb0: tail call void @g(i32 0) br label %return
bb1: tail call void @g(i32 1) br label %return
bb2: tail call void @g(i32 2) br label %return
return: ret void
; This will also be lowered as bit test, but as the range [0,8] is not fully
; covered (5 missing), the default statement can be jumped to and we end up
; with one more branch.
; CHECK-LABEL: bt_is_better2
;
; 73 = 2^0 + 2^3 + 2^6
; CHECK: movl $73
; CHECK: btl
; 146 = 2^1 + 2^4 + 2^7
; CHECK: movl $146
; CHECK: btl
; 260 = 2^2 + 2^8
; CHECK: movl $260
; CHECK: btl
}
define void @bt_is_better3(i32 %x) {
entry:
switch i32 %x, label %return [
i32 10, label %bb0
i32 13, label %bb0
i32 16, label %bb0
i32 11, label %bb1
i32 14, label %bb1
i32 17, label %bb1
i32 12, label %bb2
i32 18, label %bb2
]
bb0: tail call void @g(i32 0) br label %return
bb1: tail call void @g(i32 1) br label %return
bb2: tail call void @g(i32 2) br label %return
return: ret void
; We don't have to subtract 10 from the case value to let the range become
; [0, 8], as each value in the range [10, 18] can be represented by bits in a
; word. Then we still need a branch to jump to the default statement for the
; range [0, 10).
; CHECK-LABEL: bt_is_better3
;
; 74752 = 2^10 + 2^13 + 2^16
; CHECK: movl $74752
; CHECK: btl
; 149504 = 2^11 + 2^14 + 2^17
; CHECK: movl $149504
; CHECK: btl
; 266240 = 2^12 + 2^15 + 2^18
; CHECK: movl $266240
Re-commit r235560: Switch lowering: extract jump tables and bit tests before building binary tree (PR22262) Third time's the charm. The previous commit was reverted as a reverse for-loop in SelectionDAGBuilder::lowerWorkItem did 'I--' on an iterator at the beginning of a vector, causing asserts when using debugging iterators. This commit fixes that. llvm-svn: 235608
2015-04-24 00:45:24 +08:00
; CHECK: btl
}
define void @optimal_pivot1(i32 %x) {
entry:
switch i32 %x, label %return [
i32 100, label %bb0
i32 200, label %bb1
i32 300, label %bb0
i32 400, label %bb1
i32 500, label %bb0
i32 600, label %bb1
]
bb0: tail call void @g(i32 0) br label %return
bb1: tail call void @g(i32 1) br label %return
return: ret void
; Should pivot around 400 for two subtrees of equal size.
; CHECK-LABEL: optimal_pivot1
; CHECK-NOT: cmpl
; CHECK: cmpl $399
}
define void @optimal_pivot2(i32 %x) {
entry:
switch i32 %x, label %return [
i32 100, label %bb0 i32 101, label %bb1 i32 102, label %bb2 i32 103, label %bb3
i32 200, label %bb0 i32 201, label %bb1 i32 202, label %bb2 i32 203, label %bb3
i32 300, label %bb0 i32 301, label %bb1 i32 302, label %bb2 i32 303, label %bb3
i32 400, label %bb0 i32 401, label %bb1 i32 402, label %bb2 i32 403, label %bb3
]
bb0: tail call void @g(i32 0) br label %return
bb1: tail call void @g(i32 1) br label %return
bb2: tail call void @g(i32 2) br label %return
bb3: tail call void @g(i32 3) br label %return
return: ret void
; Should pivot around 300 for two subtrees with two jump tables each.
; CHECK-LABEL: optimal_pivot2
; CHECK-NOT: cmpl
; CHECK: cmpl $299
; CHECK: jmpq *.LJTI
; CHECK: jmpq *.LJTI
; CHECK: jmpq *.LJTI
; CHECK: jmpq *.LJTI
}
define void @optimal_jump_table1(i32 %x) {
entry:
switch i32 %x, label %return [
i32 0, label %bb0
i32 5, label %bb1
i32 6, label %bb2
i32 12, label %bb3
i32 13, label %bb4
i32 15, label %bb5
]
bb0: tail call void @g(i32 0) br label %return
bb1: tail call void @g(i32 1) br label %return
bb2: tail call void @g(i32 2) br label %return
bb3: tail call void @g(i32 3) br label %return
bb4: tail call void @g(i32 4) br label %return
bb5: tail call void @g(i32 5) br label %return
return: ret void
; Splitting in the largest gap (between 6 and 12) would yield suboptimal result.
; Expecting a jump table from 5 to 15.
; CHECK-LABEL: optimal_jump_table1
; CHECK: leal -5
; CHECK: cmpl $10
; CHECK: jmpq *.LJTI
Switch lowering: enable whole-switch jump tables at -O0. To same compile time, the analysis to find dense case-clusters in switches is not done at -O0. However, when the whole switch is dense enough, it is easy to turn it into a jump table, resulting in much faster code with no extra effort. llvm-svn: 240071
2015-06-19 06:22:30 +08:00
; At -O0, we don't build jump tables for only parts of a switch.
; NOOPT-LABEL: optimal_jump_table1
; NOOPT: testl %edi, %edi
; NOOPT: je
Relax fast register allocator related test cases; NFC - Relex hard coded registers and stack frame sizes - Some test cleanups - Change phi-dbg.ll to match on mir output after phi elimination instead of going through the whole codegen pipeline. This is in preparation for https://reviews.llvm.org/D52010 I'm committing all the test changes upfront that work before and after independently. llvm-svn: 345532
2018-10-30 04:10:42 +08:00
; NOOPT: subl $5, [[REG:%e[abcd][xi]]]
Switch lowering: enable whole-switch jump tables at -O0. To same compile time, the analysis to find dense case-clusters in switches is not done at -O0. However, when the whole switch is dense enough, it is easy to turn it into a jump table, resulting in much faster code with no extra effort. llvm-svn: 240071
2015-06-19 06:22:30 +08:00
; NOOPT: je
Relax fast register allocator related test cases; NFC - Relex hard coded registers and stack frame sizes - Some test cleanups - Change phi-dbg.ll to match on mir output after phi elimination instead of going through the whole codegen pipeline. This is in preparation for https://reviews.llvm.org/D52010 I'm committing all the test changes upfront that work before and after independently. llvm-svn: 345532
2018-10-30 04:10:42 +08:00
; NOOPT: subl $6, [[REG]]
Switch lowering: enable whole-switch jump tables at -O0. To same compile time, the analysis to find dense case-clusters in switches is not done at -O0. However, when the whole switch is dense enough, it is easy to turn it into a jump table, resulting in much faster code with no extra effort. llvm-svn: 240071
2015-06-19 06:22:30 +08:00
; NOOPT: je
Relax fast register allocator related test cases; NFC - Relex hard coded registers and stack frame sizes - Some test cleanups - Change phi-dbg.ll to match on mir output after phi elimination instead of going through the whole codegen pipeline. This is in preparation for https://reviews.llvm.org/D52010 I'm committing all the test changes upfront that work before and after independently. llvm-svn: 345532
2018-10-30 04:10:42 +08:00
; NOOPT: subl $12, [[REG]]
Switch lowering: enable whole-switch jump tables at -O0. To same compile time, the analysis to find dense case-clusters in switches is not done at -O0. However, when the whole switch is dense enough, it is easy to turn it into a jump table, resulting in much faster code with no extra effort. llvm-svn: 240071
2015-06-19 06:22:30 +08:00
; NOOPT: je
Relax fast register allocator related test cases; NFC - Relex hard coded registers and stack frame sizes - Some test cleanups - Change phi-dbg.ll to match on mir output after phi elimination instead of going through the whole codegen pipeline. This is in preparation for https://reviews.llvm.org/D52010 I'm committing all the test changes upfront that work before and after independently. llvm-svn: 345532
2018-10-30 04:10:42 +08:00
; NOOPT: subl $13, [[REG]]
Switch lowering: enable whole-switch jump tables at -O0. To same compile time, the analysis to find dense case-clusters in switches is not done at -O0. However, when the whole switch is dense enough, it is easy to turn it into a jump table, resulting in much faster code with no extra effort. llvm-svn: 240071
2015-06-19 06:22:30 +08:00
; NOOPT: je
Relax fast register allocator related test cases; NFC - Relex hard coded registers and stack frame sizes - Some test cleanups - Change phi-dbg.ll to match on mir output after phi elimination instead of going through the whole codegen pipeline. This is in preparation for https://reviews.llvm.org/D52010 I'm committing all the test changes upfront that work before and after independently. llvm-svn: 345532
2018-10-30 04:10:42 +08:00
; NOOPT: subl $15, [[REG]]
Switch lowering: enable whole-switch jump tables at -O0. To same compile time, the analysis to find dense case-clusters in switches is not done at -O0. However, when the whole switch is dense enough, it is easy to turn it into a jump table, resulting in much faster code with no extra effort. llvm-svn: 240071
2015-06-19 06:22:30 +08:00
; NOOPT: je
Re-commit r235560: Switch lowering: extract jump tables and bit tests before building binary tree (PR22262) Third time's the charm. The previous commit was reverted as a reverse for-loop in SelectionDAGBuilder::lowerWorkItem did 'I--' on an iterator at the beginning of a vector, causing asserts when using debugging iterators. This commit fixes that. llvm-svn: 235608
2015-04-24 00:45:24 +08:00
}
define void @optimal_jump_table2(i32 %x) {
entry:
switch i32 %x, label %return [
i32 0, label %bb0
i32 1, label %bb1
i32 2, label %bb2
i32 9, label %bb3
i32 14, label %bb4
i32 15, label %bb5
]
bb0: tail call void @g(i32 0) br label %return
bb1: tail call void @g(i32 1) br label %return
bb2: tail call void @g(i32 2) br label %return
bb3: tail call void @g(i32 3) br label %return
bb4: tail call void @g(i32 4) br label %return
bb5: tail call void @g(i32 5) br label %return
return: ret void
; Partitioning the cases to the minimum number of dense sets is not good enough.
; This can be partitioned as {0,1,2,9},{14,15} or {0,1,2},{9,14,15}. The former
; should be preferred. Expecting a table from 0-9.
; CHECK-LABEL: optimal_jump_table2
; CHECK: cmpl $9
; CHECK: jmpq *.LJTI
}
define void @optimal_jump_table3(i32 %x) {
entry:
switch i32 %x, label %return [
i32 1, label %bb0
i32 2, label %bb1
i32 3, label %bb2
i32 10, label %bb3
i32 13, label %bb0
i32 14, label %bb1
i32 15, label %bb2
i32 20, label %bb3
i32 25, label %bb4
]
bb0: tail call void @g(i32 0) br label %return
bb1: tail call void @g(i32 1) br label %return
bb2: tail call void @g(i32 2) br label %return
bb3: tail call void @g(i32 3) br label %return
bb4: tail call void @g(i32 4) br label %return
return: ret void
; Splitting to maximize left-right density sum and gap size would split this
; between 3 and 10, and then between 20 and 25. It's better to build a table
; from 1-20.
; CHECK-LABEL: optimal_jump_table3
; CHECK: leal -1
; CHECK: cmpl $19
; CHECK: jmpq *.LJTI
}
%struct.S = type { %struct.S*, i32 }
define void @phi_node_trouble(%struct.S* %s) {
entry:
br label %header
header:
%ptr = phi %struct.S* [ %s, %entry ], [ %next, %loop ]
%bool = icmp eq %struct.S* %ptr, null
br i1 %bool, label %exit, label %loop
loop:
%nextptr = getelementptr inbounds %struct.S, %struct.S* %ptr, i64 0, i32 0
%next = load %struct.S*, %struct.S** %nextptr
%xptr = getelementptr inbounds %struct.S, %struct.S* %next, i64 0, i32 1
%x = load i32, i32* %xptr
switch i32 %x, label %exit [
i32 4, label %header
i32 36, label %exit2
i32 69, label %exit2
i32 25, label %exit2
]
exit:
ret void
exit2:
ret void
; This will be lowered to a comparison with 4 and then bit tests. Make sure
; that the phi node in %header gets a value from the comparison block.
; CHECK-LABEL: phi_node_trouble
; CHECK: movq (%[[REG1:[a-z]+]]), %[[REG1]]
; CHECK: movl 8(%[[REG1]]), %[[REG2:[a-z]+]]
; CHECK: cmpl $4, %[[REG2]]
}
define void @default_only(i32 %x) {
entry:
br label %sw
return:
ret void
sw:
switch i32 %x, label %return [
]
; Branch directly to the default.
; (In optimized builds the switch is removed earlier.)
; NOOPT-LABEL: default_only
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; NOOPT: .LBB[[L:[A-Z0-9_]+]]:
Re-commit r235560: Switch lowering: extract jump tables and bit tests before building binary tree (PR22262) Third time's the charm. The previous commit was reverted as a reverse for-loop in SelectionDAGBuilder::lowerWorkItem did 'I--' on an iterator at the beginning of a vector, causing asserts when using debugging iterators. This commit fixes that. llvm-svn: 235608
2015-04-24 00:45:24 +08:00
; NOOPT-NEXT: retq
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; NOOPT: jmp .LBB[[L]]
Re-commit r235560: Switch lowering: extract jump tables and bit tests before building binary tree (PR22262) Third time's the charm. The previous commit was reverted as a reverse for-loop in SelectionDAGBuilder::lowerWorkItem did 'I--' on an iterator at the beginning of a vector, causing asserts when using debugging iterators. This commit fixes that. llvm-svn: 235608
2015-04-24 00:45:24 +08:00
}
Switch lowering: fix APInt overflow causing infinite loop / OOM llvm-svn: 235729
2015-04-25 00:53:55 +08:00
define void @int_max_table_cluster(i8 %x) {
entry:
switch i8 %x, label %return [
i8 0, label %bb0 i8 1, label %bb0 i8 2, label %bb0 i8 3, label %bb0
i8 4, label %bb0 i8 5, label %bb0 i8 6, label %bb0 i8 7, label %bb0
i8 8, label %bb0 i8 9, label %bb0 i8 10, label %bb0 i8 11, label %bb0
i8 12, label %bb0 i8 13, label %bb0 i8 14, label %bb0 i8 15, label %bb0
i8 16, label %bb0 i8 17, label %bb0 i8 18, label %bb0 i8 19, label %bb0
i8 20, label %bb0 i8 21, label %bb0 i8 22, label %bb0 i8 23, label %bb0
i8 24, label %bb0 i8 25, label %bb0 i8 26, label %bb0 i8 27, label %bb0
i8 28, label %bb0 i8 29, label %bb0 i8 30, label %bb0 i8 31, label %bb0
i8 32, label %bb0 i8 33, label %bb0 i8 34, label %bb0 i8 35, label %bb0
i8 36, label %bb0 i8 37, label %bb0 i8 38, label %bb0 i8 39, label %bb0
i8 40, label %bb0 i8 41, label %bb0 i8 42, label %bb0 i8 43, label %bb0
i8 44, label %bb0 i8 45, label %bb0 i8 46, label %bb0 i8 47, label %bb0
i8 48, label %bb0 i8 49, label %bb0 i8 50, label %bb0 i8 51, label %bb0
i8 52, label %bb0 i8 53, label %bb0 i8 54, label %bb0 i8 55, label %bb0
i8 56, label %bb0 i8 57, label %bb0 i8 58, label %bb0 i8 59, label %bb0
i8 60, label %bb0 i8 61, label %bb0 i8 62, label %bb0 i8 63, label %bb0
i8 64, label %bb0 i8 65, label %bb0 i8 66, label %bb0 i8 67, label %bb0
i8 68, label %bb0 i8 69, label %bb0 i8 70, label %bb0 i8 71, label %bb0
i8 72, label %bb0 i8 73, label %bb0 i8 74, label %bb0 i8 75, label %bb0
i8 76, label %bb0 i8 77, label %bb0 i8 78, label %bb0 i8 79, label %bb0
i8 80, label %bb0 i8 81, label %bb0 i8 82, label %bb0 i8 83, label %bb0
i8 84, label %bb0 i8 85, label %bb0 i8 86, label %bb0 i8 87, label %bb0
i8 88, label %bb0 i8 89, label %bb0 i8 90, label %bb0 i8 91, label %bb0
i8 92, label %bb0 i8 93, label %bb0 i8 94, label %bb0 i8 95, label %bb0
i8 96, label %bb0 i8 97, label %bb0 i8 98, label %bb0 i8 99, label %bb0
i8 100, label %bb0 i8 101, label %bb0 i8 102, label %bb0 i8 103, label %bb0
i8 104, label %bb0 i8 105, label %bb0 i8 106, label %bb0 i8 107, label %bb0
i8 108, label %bb0 i8 109, label %bb0 i8 110, label %bb0 i8 111, label %bb0
i8 112, label %bb0 i8 113, label %bb0 i8 114, label %bb0 i8 115, label %bb0
i8 116, label %bb0 i8 117, label %bb0 i8 118, label %bb0 i8 119, label %bb0
i8 120, label %bb0 i8 121, label %bb0 i8 122, label %bb0 i8 123, label %bb0
i8 124, label %bb0 i8 125, label %bb0 i8 126, label %bb0 i8 127, label %bb0
i8 -64, label %bb1 i8 -63, label %bb1 i8 -62, label %bb1 i8 -61, label %bb1
i8 -60, label %bb1 i8 -59, label %bb1 i8 -58, label %bb1 i8 -57, label %bb1
i8 -56, label %bb1 i8 -55, label %bb1 i8 -54, label %bb1 i8 -53, label %bb1
i8 -52, label %bb1 i8 -51, label %bb1 i8 -50, label %bb1 i8 -49, label %bb1
i8 -48, label %bb1 i8 -47, label %bb1 i8 -46, label %bb1 i8 -45, label %bb1
i8 -44, label %bb1 i8 -43, label %bb1 i8 -42, label %bb1 i8 -41, label %bb1
i8 -40, label %bb1 i8 -39, label %bb1 i8 -38, label %bb1 i8 -37, label %bb1
i8 -36, label %bb1 i8 -35, label %bb1 i8 -34, label %bb1 i8 -33, label %bb1
i8 -32, label %bb2 i8 -31, label %bb2 i8 -30, label %bb2 i8 -29, label %bb2
i8 -28, label %bb2 i8 -27, label %bb2 i8 -26, label %bb2 i8 -25, label %bb2
i8 -24, label %bb2 i8 -23, label %bb2 i8 -22, label %bb2 i8 -21, label %bb2
i8 -20, label %bb2 i8 -19, label %bb2 i8 -18, label %bb2 i8 -17, label %bb2
i8 -16, label %bb3 i8 -15, label %bb3 i8 -14, label %bb3 i8 -13, label %bb3
i8 -12, label %bb3 i8 -11, label %bb3 i8 -10, label %bb3 i8 -9, label %bb3
]
bb0: tail call void @g(i32 0) br label %return
bb1: tail call void @g(i32 1) br label %return
bb2: tail call void @g(i32 1) br label %return
bb3: tail call void @g(i32 1) br label %return
return: ret void
; Don't infloop on jump tables where the upper bound is the max value of the
; input type (in this case 127).
; CHECK-LABEL: int_max_table_cluster
; CHECK: jmpq *.LJTI
}
Switch lowering: order bit tests by branch weight. llvm-svn: 235912
2015-04-28 04:21:17 +08:00
define void @bt_order_by_weight(i32 %x) {
entry:
switch i32 %x, label %return [
i32 0, label %bb0
i32 3, label %bb0
i32 6, label %bb0
i32 1, label %bb1
i32 4, label %bb1
i32 7, label %bb1
i32 2, label %bb2
i32 5, label %bb2
i32 8, label %bb2
i32 9, label %bb2
], !prof !1
bb0: tail call void @g(i32 0) br label %return
bb1: tail call void @g(i32 1) br label %return
bb2: tail call void @g(i32 2) br label %return
return: ret void
; Cases 1,4,7 have a very large branch weight (which shouldn't overflow), so
; their bit test should come first. 0,3,6 and 2,5,8,9 both have a weight of 12,
; but the latter set has more cases, so should be tested for earlier.
Remove the final bit test during lowering switch statement if all cases in bit test cover a contiguous range. When lowering switch statement, if bit tests are used then LLVM will always generates a jump to the default statement in the last bit test. However, this is not necessary when all cases in bit tests cover a contiguous range. This is because when generating the bit tests header MBB, there is a range check that guarantees cases in bit tests won't go outside of [low, high], where low and high are minimum and maximum case values in the bit tests. This patch checks if this is the case and then doesn't emit jump to default statement and hence saves a bit test and a branch. Differential Revision: http://reviews.llvm.org/D12249 llvm-svn: 245976
2015-08-26 05:34:38 +08:00
; The bit test on 0,3,6 is unnecessary as all cases cover the rage [0, 9].
; The range check guarantees that cases other than 1,4,7 and 2,5,8,9 must be
; in 0,3,6.
Switch lowering: order bit tests by branch weight. llvm-svn: 235912
2015-04-28 04:21:17 +08:00
; CHECK-LABEL: bt_order_by_weight
; 146 = 2^1 + 2^4 + 2^7
; CHECK: movl $146
; CHECK: btl
; 292 = 2^2 + 2^5 + 2^8 + 2^9
; CHECK: movl $804
; CHECK: btl
; 73 = 2^0 + 2^3 + 2^6
Remove the final bit test during lowering switch statement if all cases in bit test cover a contiguous range. When lowering switch statement, if bit tests are used then LLVM will always generates a jump to the default statement in the last bit test. However, this is not necessary when all cases in bit tests cover a contiguous range. This is because when generating the bit tests header MBB, there is a range check that guarantees cases in bit tests won't go outside of [low, high], where low and high are minimum and maximum case values in the bit tests. This patch checks if this is the case and then doesn't emit jump to default statement and hence saves a bit test and a branch. Differential Revision: http://reviews.llvm.org/D12249 llvm-svn: 245976
2015-08-26 05:34:38 +08:00
; CHECK-NOT: movl $73
; CHECK-NOT: btl
Switch lowering: order bit tests by branch weight. llvm-svn: 235912
2015-04-28 04:21:17 +08:00
}
!1 = !{!"branch_weights",
; Default:
i32 1,
; Cases 0,3,6:
i32 4, i32 4, i32 4,
; Cases 1,4,7:
i32 4294967295, i32 2, i32 4294967295,
; Cases 2,5,8,9:
i32 3, i32 3, i32 3, i32 3}
Switch lowering: Take branch weight into account when ordering for fall-through Previously, the code would try to put a fall-through case last, even if that meant moving a case with much higher branch weight further down the chain. Ordering by branch weight is most important, putting a fall-through block last is secondary. llvm-svn: 235942
2015-04-28 07:35:22 +08:00
define void @order_by_weight_and_fallthrough(i32 %x) {
entry:
switch i32 %x, label %return [
i32 100, label %bb1
i32 200, label %bb0
i32 300, label %bb0
], !prof !2
bb0: tail call void @g(i32 0) br label %return
bb1: tail call void @g(i32 1) br label %return
return: ret void
; Case 200 has the highest weight and should come first. 100 and 300 have the
; same weight, but 300 goes to the 'next' block, so should be last.
; CHECK-LABEL: order_by_weight_and_fallthrough
; CHECK: cmpl $200
; CHECK: cmpl $100
; CHECK: cmpl $300
}
!2 = !{!"branch_weights",
; Default:
i32 1,
; Case 100:
i32 10,
; Case 200:
i32 1000,
; Case 300:
i32 10}
Switch lowering: use profile info to build weight-balanced binary search trees This will cause hot nodes to appear closer to the root. The literature says building the tree like this makes it a near-optimal (in terms of search time given key frequencies) binary search tree. In LLVM's case, we can do up to 3 comparisons in each leaf node, so it might be better to opt for lower tree height in some cases; that's something to look into in the future. Differential Revision: http://reviews.llvm.org/D9318 llvm-svn: 236192
2015-04-30 08:57:37 +08:00
define void @zero_weight_tree(i32 %x) {
entry:
switch i32 %x, label %return [
i32 0, label %bb0
i32 10, label %bb1
i32 20, label %bb2
i32 30, label %bb3
i32 40, label %bb4
i32 50, label %bb5
], !prof !3
bb0: tail call void @g(i32 0) br label %return
bb1: tail call void @g(i32 1) br label %return
bb2: tail call void @g(i32 2) br label %return
bb3: tail call void @g(i32 3) br label %return
bb4: tail call void @g(i32 4) br label %return
bb5: tail call void @g(i32 5) br label %return
return: ret void
; Make sure to pick a pivot in the middle also with zero-weight cases.
; CHECK-LABEL: zero_weight_tree
; CHECK-NOT: cmpl
; CHECK: cmpl $29
}
!3 = !{!"branch_weights", i32 1, i32 10, i32 0, i32 0, i32 0, i32 0, i32 10}
define void @left_leaning_weight_balanced_tree(i32 %x) {
entry:
switch i32 %x, label %return [
i32 0, label %bb0
i32 10, label %bb1
i32 20, label %bb2
i32 30, label %bb3
i32 40, label %bb4
i32 50, label %bb5
Switch lowering: add heuristic for filling leaf nodes in the weight-balanced binary search tree Sparse switches with profile info are lowered as weight-balanced BSTs. For example, if the node weights are {1,1,1,1,1,1000}, the right-most node would end up in a tree by itself, bringing it closer to the top. However, a leaf in this BST can contain up to 3 cases, and having a single case in a leaf node as in the example means the tree might become unnecessarily high. This patch adds a heauristic to the pivot selection algorithm that moves more cases into leaf nodes unless that would lower their rank. It still doesn't yield the optimal tree in every case, but I believe it's conservatibely correct. llvm-svn: 240224
2015-06-21 01:14:07 +08:00
i32 60, label %bb6
i32 70, label %bb6
Switch lowering: use profile info to build weight-balanced binary search trees This will cause hot nodes to appear closer to the root. The literature says building the tree like this makes it a near-optimal (in terms of search time given key frequencies) binary search tree. In LLVM's case, we can do up to 3 comparisons in each leaf node, so it might be better to opt for lower tree height in some cases; that's something to look into in the future. Differential Revision: http://reviews.llvm.org/D9318 llvm-svn: 236192
2015-04-30 08:57:37 +08:00
], !prof !4
bb0: tail call void @g(i32 0) br label %return
bb1: tail call void @g(i32 1) br label %return
bb2: tail call void @g(i32 2) br label %return
bb3: tail call void @g(i32 3) br label %return
bb4: tail call void @g(i32 4) br label %return
bb5: tail call void @g(i32 5) br label %return
Switch lowering: add heuristic for filling leaf nodes in the weight-balanced binary search tree Sparse switches with profile info are lowered as weight-balanced BSTs. For example, if the node weights are {1,1,1,1,1,1000}, the right-most node would end up in a tree by itself, bringing it closer to the top. However, a leaf in this BST can contain up to 3 cases, and having a single case in a leaf node as in the example means the tree might become unnecessarily high. This patch adds a heauristic to the pivot selection algorithm that moves more cases into leaf nodes unless that would lower their rank. It still doesn't yield the optimal tree in every case, but I believe it's conservatibely correct. llvm-svn: 240224
2015-06-21 01:14:07 +08:00
bb6: tail call void @g(i32 6) br label %return
bb7: tail call void @g(i32 7) br label %return
Switch lowering: use profile info to build weight-balanced binary search trees This will cause hot nodes to appear closer to the root. The literature says building the tree like this makes it a near-optimal (in terms of search time given key frequencies) binary search tree. In LLVM's case, we can do up to 3 comparisons in each leaf node, so it might be better to opt for lower tree height in some cases; that's something to look into in the future. Differential Revision: http://reviews.llvm.org/D9318 llvm-svn: 236192
2015-04-30 08:57:37 +08:00
return: ret void
Switch lowering: add heuristic for filling leaf nodes in the weight-balanced binary search tree Sparse switches with profile info are lowered as weight-balanced BSTs. For example, if the node weights are {1,1,1,1,1,1000}, the right-most node would end up in a tree by itself, bringing it closer to the top. However, a leaf in this BST can contain up to 3 cases, and having a single case in a leaf node as in the example means the tree might become unnecessarily high. This patch adds a heauristic to the pivot selection algorithm that moves more cases into leaf nodes unless that would lower their rank. It still doesn't yield the optimal tree in every case, but I believe it's conservatibely correct. llvm-svn: 240224
2015-06-21 01:14:07 +08:00
; Without branch probabilities, the pivot would be 40, since that would yield
; equal-sized sub-trees. When taking weights into account, case 70 becomes the
; pivot. Since there is room for 3 cases in a leaf, cases 50 and 60 are also
; included in the right-hand side because that doesn't reduce their rank.
Switch lowering: use profile info to build weight-balanced binary search trees This will cause hot nodes to appear closer to the root. The literature says building the tree like this makes it a near-optimal (in terms of search time given key frequencies) binary search tree. In LLVM's case, we can do up to 3 comparisons in each leaf node, so it might be better to opt for lower tree height in some cases; that's something to look into in the future. Differential Revision: http://reviews.llvm.org/D9318 llvm-svn: 236192
2015-04-30 08:57:37 +08:00
; CHECK-LABEL: left_leaning_weight_balanced_tree
; CHECK-NOT: cmpl
Switch lowering: add heuristic for filling leaf nodes in the weight-balanced binary search tree Sparse switches with profile info are lowered as weight-balanced BSTs. For example, if the node weights are {1,1,1,1,1,1000}, the right-most node would end up in a tree by itself, bringing it closer to the top. However, a leaf in this BST can contain up to 3 cases, and having a single case in a leaf node as in the example means the tree might become unnecessarily high. This patch adds a heauristic to the pivot selection algorithm that moves more cases into leaf nodes unless that would lower their rank. It still doesn't yield the optimal tree in every case, but I believe it's conservatibely correct. llvm-svn: 240224
2015-06-21 01:14:07 +08:00
; CHECK: cmpl $49
}
!4 = !{!"branch_weights", i32 1, i32 10, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1000}
define void @left_leaning_weight_balanced_tree2(i32 %x) {
entry:
switch i32 %x, label %return [
i32 0, label %bb0
i32 10, label %bb1
i32 20, label %bb2
i32 30, label %bb3
i32 40, label %bb4
i32 50, label %bb5
i32 60, label %bb6
i32 70, label %bb6
], !prof !5
bb0: tail call void @g(i32 0) br label %return
bb1: tail call void @g(i32 1) br label %return
bb2: tail call void @g(i32 2) br label %return
bb3: tail call void @g(i32 3) br label %return
bb4: tail call void @g(i32 4) br label %return
bb5: tail call void @g(i32 5) br label %return
bb6: tail call void @g(i32 6) br label %return
bb7: tail call void @g(i32 7) br label %return
return: ret void
; Same as the previous test, except case 50 has higher rank to the left than it
; would have on the right. Case 60 would have the same rank on both sides, so is
; moved into the leaf.
; CHECK-LABEL: left_leaning_weight_balanced_tree2
; CHECK-NOT: cmpl
; CHECK: cmpl $59
}
!5 = !{!"branch_weights", i32 1, i32 10, i32 1, i32 1, i32 1, i32 1, i32 90, i32 70, i32 1000}
define void @right_leaning_weight_balanced_tree(i32 %x) {
entry:
switch i32 %x, label %return [
i32 0, label %bb0
i32 10, label %bb1
i32 20, label %bb2
i32 30, label %bb3
i32 40, label %bb4
i32 50, label %bb5
i32 60, label %bb6
i32 70, label %bb6
], !prof !6
bb0: tail call void @g(i32 0) br label %return
bb1: tail call void @g(i32 1) br label %return
bb2: tail call void @g(i32 2) br label %return
bb3: tail call void @g(i32 3) br label %return
bb4: tail call void @g(i32 4) br label %return
bb5: tail call void @g(i32 5) br label %return
bb6: tail call void @g(i32 6) br label %return
bb7: tail call void @g(i32 7) br label %return
return: ret void
; Analogous to left_leaning_weight_balanced_tree.
; CHECK-LABEL: right_leaning_weight_balanced_tree
; CHECK-NOT: cmpl
; CHECK: cmpl $19
Switch lowering: use profile info to build weight-balanced binary search trees This will cause hot nodes to appear closer to the root. The literature says building the tree like this makes it a near-optimal (in terms of search time given key frequencies) binary search tree. In LLVM's case, we can do up to 3 comparisons in each leaf node, so it might be better to opt for lower tree height in some cases; that's something to look into in the future. Differential Revision: http://reviews.llvm.org/D9318 llvm-svn: 236192
2015-04-30 08:57:37 +08:00
}
Switch lowering: add heuristic for filling leaf nodes in the weight-balanced binary search tree Sparse switches with profile info are lowered as weight-balanced BSTs. For example, if the node weights are {1,1,1,1,1,1000}, the right-most node would end up in a tree by itself, bringing it closer to the top. However, a leaf in this BST can contain up to 3 cases, and having a single case in a leaf node as in the example means the tree might become unnecessarily high. This patch adds a heauristic to the pivot selection algorithm that moves more cases into leaf nodes unless that would lower their rank. It still doesn't yield the optimal tree in every case, but I believe it's conservatibely correct. llvm-svn: 240224
2015-06-21 01:14:07 +08:00
!6 = !{!"branch_weights", i32 1, i32 1000, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 10}
Switch lowering: use profile info to build weight-balanced binary search trees This will cause hot nodes to appear closer to the root. The literature says building the tree like this makes it a near-optimal (in terms of search time given key frequencies) binary search tree. In LLVM's case, we can do up to 3 comparisons in each leaf node, so it might be better to opt for lower tree height in some cases; that's something to look into in the future. Differential Revision: http://reviews.llvm.org/D9318 llvm-svn: 236192
2015-04-30 08:57:37 +08:00
define void @jump_table_affects_balance(i32 %x) {
entry:
switch i32 %x, label %return [
; Jump table:
i32 0, label %bb0
i32 1, label %bb1
i32 2, label %bb2
i32 3, label %bb3
i32 100, label %bb0
i32 200, label %bb1
i32 300, label %bb2
]
bb0: tail call void @g(i32 0) br label %return
bb1: tail call void @g(i32 1) br label %return
bb2: tail call void @g(i32 2) br label %return
bb3: tail call void @g(i32 3) br label %return
return: ret void
; CHECK-LABEL: jump_table_affects_balance
; If the tree were balanced based on number of clusters, {0-3,100} would go on
; the left and {200,300} on the right. However, the jump table weights as much
; as its components, so 100 is selected as the pivot.
; CHECK-NOT: cmpl
; CHECK: cmpl $99
}
Switch lowering: fix assert in buildBitTests (PR23738) When checking (High - Low + 1).sle(BitWidth), BitWidth would be truncated to the size of the left-hand side. In the case of this PR, the left-hand side was i4, so BitWidth=64 got truncated to 0 and the assert failed. llvm-svn: 239048
2015-06-04 23:55:00 +08:00
define void @pr23738(i4 %x) {
entry:
switch i4 %x, label %bb0 [
i4 0, label %bb1
i4 1, label %bb1
i4 -5, label %bb1
]
bb0: tail call void @g(i32 0) br label %return
bb1: tail call void @g(i32 1) br label %return
return: ret void
; Don't assert due to truncating the bitwidth (64) to i4 when checking
; that the bit-test range fits in a word.
}
Switch lowering: don't add incoming PHI values from skipped bit test MBB's (PR27135) After r245976, LLVM will skip the last bit test case if knows it will always be true. However, we would still erroneously update PHI nodes with incoming values from the MBB that would perform the final bit test, causing -verify-machineinstrs to fail. llvm-svn: 266479
2016-04-16 05:45:30 +08:00
The associated PR for this test was PR27135 not PR27132. llvm-svn: 268772
2016-05-07 02:23:14 +08:00
define i32 @pr27135(i32 %i) {
Switch lowering: don't add incoming PHI values from skipped bit test MBB's (PR27135) After r245976, LLVM will skip the last bit test case if knows it will always be true. However, we would still erroneously update PHI nodes with incoming values from the MBB that would perform the final bit test, causing -verify-machineinstrs to fail. llvm-svn: 266479
2016-04-16 05:45:30 +08:00
entry:
br i1 undef, label %sw, label %end
sw:
switch i32 %i, label %end [
i32 99, label %sw.bb
i32 98, label %sw.bb
i32 101, label %sw.bb
i32 97, label %sw.bb2
i32 96, label %sw.bb2
i32 100, label %sw.bb2
]
sw.bb:
unreachable
sw.bb2:
unreachable
end:
%p = phi i32 [ 1, %sw ], [ 0, %entry ]
ret i32 %p
The associated PR for this test was PR27135 not PR27132. llvm-svn: 268772
2016-05-07 02:23:14 +08:00
; CHECK-LABEL: pr27135:
Switch lowering: don't add incoming PHI values from skipped bit test MBB's (PR27135) After r245976, LLVM will skip the last bit test case if knows it will always be true. However, we would still erroneously update PHI nodes with incoming values from the MBB that would perform the final bit test, causing -verify-machineinstrs to fail. llvm-svn: 266479
2016-04-16 05:45:30 +08:00
; The switch is lowered with bit tests. Since the case range is contiguous, the
; second bit test is redundant and can be skipped. Check that we don't update
; the phi node with an incoming value from the MBB of the skipped bit test
; (-verify-machine-instrs cathces this).
; CHECK: btl
; CHECK-NOT: btl
}