maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/Transforms/RewriteStatepointsForGC/base-vector.ll

248 lines
10 KiB
LLVM
Raw Normal View History

[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
; RUN: opt < %s -rewrite-statepoints-for-gc -S | FileCheck %s
[RewriteStatepointsForGC] Handle extractelement fully in the base pointer algorithm When rewriting the IR such that base pointers are available for every live pointer, we potentially need to duplicate instructions to propagate the base. The original code had only handled PHI and Select under the belief those were the only instructions which would need duplicated. When I added support for vector instructions, I'd added a collection of hacks for ExtractElement which caught most of the common cases. Of course, I then found the one test case my hacks couldn't cover. :) This change removes all of the early hacks for extract element. By defining extractelement as a BDV (rather than trying to look through it), we can extend the rewriting algorithm to duplicate the extract as needed. Note that a couple of peephole optimizations were left in for the moment, because while we now handle extractelement as a first class citizen, we're not yet handling insertelement. That change will follow in the near future. llvm-svn: 244808
2015-08-13 05:00:20 +08:00
define i64 addrspace(1)* @test(<2 x i64 addrspace(1)*> %vec, i32 %idx) gc "statepoint-example" {
; CHECK-LABEL: @test
; CHECK: extractelement
; CHECK: extractelement
; CHECK: statepoint
; CHECK: gc.relocate
; CHECK-DAG: ; (%base_ee, %obj)
[RS4GC] Use SetVector/MapVector instead of DenseSet/DenseMap to guarantee stable ordering Goal of this change is to guarantee stable ordering of the statepoint arguments and other newly inserted values such as gc.relocates. Previously we had explicit sorting in a couple of places. However for unnamed values ordering was partial and overall we didn't have any strong invariant regarding it. This change switches all data structures to use SetVector's and MapVector's which provide possibility for deterministic iteration over them. Explicit sorting is now redundant and was removed. Differential Revision: http://reviews.llvm.org/D19669 llvm-svn: 268502
2016-05-04 22:55:36 +08:00
; CHECK: gc.relocate
; CHECK-DAG: ; (%base_ee, %base_ee)
[RewriteStatepointsForGC] Handle extractelement fully in the base pointer algorithm When rewriting the IR such that base pointers are available for every live pointer, we potentially need to duplicate instructions to propagate the base. The original code had only handled PHI and Select under the belief those were the only instructions which would need duplicated. When I added support for vector instructions, I'd added a collection of hacks for ExtractElement which caught most of the common cases. Of course, I then found the one test case my hacks couldn't cover. :) This change removes all of the early hacks for extract element. By defining extractelement as a BDV (rather than trying to look through it), we can extend the rewriting algorithm to duplicate the extract as needed. Note that a couple of peephole optimizations were left in for the moment, because while we now handle extractelement as a first class citizen, we're not yet handling insertelement. That change will follow in the near future. llvm-svn: 244808
2015-08-13 05:00:20 +08:00
; Note that the second extractelement is actually redundant here. A correct output would
; be to reuse the existing obj as a base since it is actually a base pointer.
entry:
%obj = extractelement <2 x i64 addrspace(1)*> %vec, i32 %idx
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
call void @do_safepoint() [ "deopt"() ]
[RewriteStatepointsForGC] Handle extractelement fully in the base pointer algorithm When rewriting the IR such that base pointers are available for every live pointer, we potentially need to duplicate instructions to propagate the base. The original code had only handled PHI and Select under the belief those were the only instructions which would need duplicated. When I added support for vector instructions, I'd added a collection of hacks for ExtractElement which caught most of the common cases. Of course, I then found the one test case my hacks couldn't cover. :) This change removes all of the early hacks for extract element. By defining extractelement as a BDV (rather than trying to look through it), we can extend the rewriting algorithm to duplicate the extract as needed. Note that a couple of peephole optimizations were left in for the moment, because while we now handle extractelement as a first class citizen, we're not yet handling insertelement. That change will follow in the near future. llvm-svn: 244808
2015-08-13 05:00:20 +08:00
ret i64 addrspace(1)* %obj
}
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
define i64 addrspace(1)* @test2(<2 x i64 addrspace(1)*>* %ptr, i1 %cnd, i32 %idx1, i32 %idx2) gc "statepoint-example" {
[RewriteStatepointsForGC] Handle extractelement fully in the base pointer algorithm When rewriting the IR such that base pointers are available for every live pointer, we potentially need to duplicate instructions to propagate the base. The original code had only handled PHI and Select under the belief those were the only instructions which would need duplicated. When I added support for vector instructions, I'd added a collection of hacks for ExtractElement which caught most of the common cases. Of course, I then found the one test case my hacks couldn't cover. :) This change removes all of the early hacks for extract element. By defining extractelement as a BDV (rather than trying to look through it), we can extend the rewriting algorithm to duplicate the extract as needed. Note that a couple of peephole optimizations were left in for the moment, because while we now handle extractelement as a first class citizen, we're not yet handling insertelement. That change will follow in the near future. llvm-svn: 244808
2015-08-13 05:00:20 +08:00
; CHECK-LABEL: test2
entry:
br i1 %cnd, label %taken, label %untaken
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
taken: ; preds = %entry
[RewriteStatepointsForGC] Handle extractelement fully in the base pointer algorithm When rewriting the IR such that base pointers are available for every live pointer, we potentially need to duplicate instructions to propagate the base. The original code had only handled PHI and Select under the belief those were the only instructions which would need duplicated. When I added support for vector instructions, I'd added a collection of hacks for ExtractElement which caught most of the common cases. Of course, I then found the one test case my hacks couldn't cover. :) This change removes all of the early hacks for extract element. By defining extractelement as a BDV (rather than trying to look through it), we can extend the rewriting algorithm to duplicate the extract as needed. Note that a couple of peephole optimizations were left in for the moment, because while we now handle extractelement as a first class citizen, we're not yet handling insertelement. That change will follow in the near future. llvm-svn: 244808
2015-08-13 05:00:20 +08:00
%obja = load <2 x i64 addrspace(1)*>, <2 x i64 addrspace(1)*>* %ptr
br label %merge
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
untaken: ; preds = %entry
[RewriteStatepointsForGC] Handle extractelement fully in the base pointer algorithm When rewriting the IR such that base pointers are available for every live pointer, we potentially need to duplicate instructions to propagate the base. The original code had only handled PHI and Select under the belief those were the only instructions which would need duplicated. When I added support for vector instructions, I'd added a collection of hacks for ExtractElement which caught most of the common cases. Of course, I then found the one test case my hacks couldn't cover. :) This change removes all of the early hacks for extract element. By defining extractelement as a BDV (rather than trying to look through it), we can extend the rewriting algorithm to duplicate the extract as needed. Note that a couple of peephole optimizations were left in for the moment, because while we now handle extractelement as a first class citizen, we're not yet handling insertelement. That change will follow in the near future. llvm-svn: 244808
2015-08-13 05:00:20 +08:00
%objb = load <2 x i64 addrspace(1)*>, <2 x i64 addrspace(1)*>* %ptr
br label %merge
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
merge: ; preds = %untaken, %taken
%vec = phi <2 x i64 addrspace(1)*> [ %obja, %taken ], [ %objb, %untaken ]
[RewriteStatepointsForGC] Handle extractelement fully in the base pointer algorithm When rewriting the IR such that base pointers are available for every live pointer, we potentially need to duplicate instructions to propagate the base. The original code had only handled PHI and Select under the belief those were the only instructions which would need duplicated. When I added support for vector instructions, I'd added a collection of hacks for ExtractElement which caught most of the common cases. Of course, I then found the one test case my hacks couldn't cover. :) This change removes all of the early hacks for extract element. By defining extractelement as a BDV (rather than trying to look through it), we can extend the rewriting algorithm to duplicate the extract as needed. Note that a couple of peephole optimizations were left in for the moment, because while we now handle extractelement as a first class citizen, we're not yet handling insertelement. That change will follow in the near future. llvm-svn: 244808
2015-08-13 05:00:20 +08:00
br i1 %cnd, label %taken2, label %untaken2
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
taken2: ; preds = %merge
[RewriteStatepointsForGC] Handle extractelement fully in the base pointer algorithm When rewriting the IR such that base pointers are available for every live pointer, we potentially need to duplicate instructions to propagate the base. The original code had only handled PHI and Select under the belief those were the only instructions which would need duplicated. When I added support for vector instructions, I'd added a collection of hacks for ExtractElement which caught most of the common cases. Of course, I then found the one test case my hacks couldn't cover. :) This change removes all of the early hacks for extract element. By defining extractelement as a BDV (rather than trying to look through it), we can extend the rewriting algorithm to duplicate the extract as needed. Note that a couple of peephole optimizations were left in for the moment, because while we now handle extractelement as a first class citizen, we're not yet handling insertelement. That change will follow in the near future. llvm-svn: 244808
2015-08-13 05:00:20 +08:00
%obj0 = extractelement <2 x i64 addrspace(1)*> %vec, i32 %idx1
br label %merge2
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
untaken2: ; preds = %merge
[RewriteStatepointsForGC] Handle extractelement fully in the base pointer algorithm When rewriting the IR such that base pointers are available for every live pointer, we potentially need to duplicate instructions to propagate the base. The original code had only handled PHI and Select under the belief those were the only instructions which would need duplicated. When I added support for vector instructions, I'd added a collection of hacks for ExtractElement which caught most of the common cases. Of course, I then found the one test case my hacks couldn't cover. :) This change removes all of the early hacks for extract element. By defining extractelement as a BDV (rather than trying to look through it), we can extend the rewriting algorithm to duplicate the extract as needed. Note that a couple of peephole optimizations were left in for the moment, because while we now handle extractelement as a first class citizen, we're not yet handling insertelement. That change will follow in the near future. llvm-svn: 244808
2015-08-13 05:00:20 +08:00
%obj1 = extractelement <2 x i64 addrspace(1)*> %vec, i32 %idx2
br label %merge2
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
merge2: ; preds = %untaken2, %taken2
[RewriteStatepointsForGC] Handle extractelement fully in the base pointer algorithm When rewriting the IR such that base pointers are available for every live pointer, we potentially need to duplicate instructions to propagate the base. The original code had only handled PHI and Select under the belief those were the only instructions which would need duplicated. When I added support for vector instructions, I'd added a collection of hacks for ExtractElement which caught most of the common cases. Of course, I then found the one test case my hacks couldn't cover. :) This change removes all of the early hacks for extract element. By defining extractelement as a BDV (rather than trying to look through it), we can extend the rewriting algorithm to duplicate the extract as needed. Note that a couple of peephole optimizations were left in for the moment, because while we now handle extractelement as a first class citizen, we're not yet handling insertelement. That change will follow in the near future. llvm-svn: 244808
2015-08-13 05:00:20 +08:00
; CHECK-LABEL: merge2:
[RS4GC] Revert optimization attempt due to memory corruption This change reverts "246133 [RewriteStatepointsForGC] Reduce the number of new instructions for base pointers" and a follow on bugfix 12575. As pointed out in pr25846, this code suffers from a memory corruption bug. Since I'm (empirically) not going to get back to this any time soon, simply reverting the problematic change is the right answer. llvm-svn: 261565
2016-02-23 04:45:56 +08:00
; CHECK: %obj.base = phi i64 addrspace(1)*
; CHECK: %obj = phi i64 addrspace(1)*
; CHECK: statepoint
; CHECK: gc.relocate
; CHECK-DAG: ; (%obj.base, %obj)
[RewriteStatepointsForGC] Handle extractelement fully in the base pointer algorithm When rewriting the IR such that base pointers are available for every live pointer, we potentially need to duplicate instructions to propagate the base. The original code had only handled PHI and Select under the belief those were the only instructions which would need duplicated. When I added support for vector instructions, I'd added a collection of hacks for ExtractElement which caught most of the common cases. Of course, I then found the one test case my hacks couldn't cover. :) This change removes all of the early hacks for extract element. By defining extractelement as a BDV (rather than trying to look through it), we can extend the rewriting algorithm to duplicate the extract as needed. Note that a couple of peephole optimizations were left in for the moment, because while we now handle extractelement as a first class citizen, we're not yet handling insertelement. That change will follow in the near future. llvm-svn: 244808
2015-08-13 05:00:20 +08:00
; CHECK: gc.relocate
[RS4GC] Revert optimization attempt due to memory corruption This change reverts "246133 [RewriteStatepointsForGC] Reduce the number of new instructions for base pointers" and a follow on bugfix 12575. As pointed out in pr25846, this code suffers from a memory corruption bug. Since I'm (empirically) not going to get back to this any time soon, simply reverting the problematic change is the right answer. llvm-svn: 261565
2016-02-23 04:45:56 +08:00
; CHECK-DAG: ; (%obj.base, %obj.base)
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
%obj = phi i64 addrspace(1)* [ %obj0, %taken2 ], [ %obj1, %untaken2 ]
call void @do_safepoint() [ "deopt"() ]
[RewriteStatepointsForGC] Handle extractelement fully in the base pointer algorithm When rewriting the IR such that base pointers are available for every live pointer, we potentially need to duplicate instructions to propagate the base. The original code had only handled PHI and Select under the belief those were the only instructions which would need duplicated. When I added support for vector instructions, I'd added a collection of hacks for ExtractElement which caught most of the common cases. Of course, I then found the one test case my hacks couldn't cover. :) This change removes all of the early hacks for extract element. By defining extractelement as a BDV (rather than trying to look through it), we can extend the rewriting algorithm to duplicate the extract as needed. Note that a couple of peephole optimizations were left in for the moment, because while we now handle extractelement as a first class citizen, we're not yet handling insertelement. That change will follow in the near future. llvm-svn: 244808
2015-08-13 05:00:20 +08:00
ret i64 addrspace(1)* %obj
}
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
define i64 addrspace(1)* @test3(i64 addrspace(1)* %ptr) gc "statepoint-example" {
[RewriteStatepointsForGC] Handle extractelement fully in the base pointer algorithm When rewriting the IR such that base pointers are available for every live pointer, we potentially need to duplicate instructions to propagate the base. The original code had only handled PHI and Select under the belief those were the only instructions which would need duplicated. When I added support for vector instructions, I'd added a collection of hacks for ExtractElement which caught most of the common cases. Of course, I then found the one test case my hacks couldn't cover. :) This change removes all of the early hacks for extract element. By defining extractelement as a BDV (rather than trying to look through it), we can extend the rewriting algorithm to duplicate the extract as needed. Note that a couple of peephole optimizations were left in for the moment, because while we now handle extractelement as a first class citizen, we're not yet handling insertelement. That change will follow in the near future. llvm-svn: 244808
2015-08-13 05:00:20 +08:00
; CHECK-LABEL: test3
; CHECK: insertelement
; CHECK: extractelement
; CHECK: statepoint
; CHECK: gc.relocate
[RS4GC] Revert optimization attempt due to memory corruption This change reverts "246133 [RewriteStatepointsForGC] Reduce the number of new instructions for base pointers" and a follow on bugfix 12575. As pointed out in pr25846, this code suffers from a memory corruption bug. Since I'm (empirically) not going to get back to this any time soon, simply reverting the problematic change is the right answer. llvm-svn: 261565
2016-02-23 04:45:56 +08:00
; CHECK-DAG: (%obj.base, %obj)
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
entry:
%vec = insertelement <2 x i64 addrspace(1)*> undef, i64 addrspace(1)* %ptr, i32 0
%obj = extractelement <2 x i64 addrspace(1)*> %vec, i32 0
call void @do_safepoint() [ "deopt"() ]
[RewriteStatepointsForGC] Handle extractelement fully in the base pointer algorithm When rewriting the IR such that base pointers are available for every live pointer, we potentially need to duplicate instructions to propagate the base. The original code had only handled PHI and Select under the belief those were the only instructions which would need duplicated. When I added support for vector instructions, I'd added a collection of hacks for ExtractElement which caught most of the common cases. Of course, I then found the one test case my hacks couldn't cover. :) This change removes all of the early hacks for extract element. By defining extractelement as a BDV (rather than trying to look through it), we can extend the rewriting algorithm to duplicate the extract as needed. Note that a couple of peephole optimizations were left in for the moment, because while we now handle extractelement as a first class citizen, we're not yet handling insertelement. That change will follow in the near future. llvm-svn: 244808
2015-08-13 05:00:20 +08:00
ret i64 addrspace(1)* %obj
}
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
define i64 addrspace(1)* @test4(i64 addrspace(1)* %ptr) gc "statepoint-example" {
[RewriteStatepointsForGC] Handle extractelement fully in the base pointer algorithm When rewriting the IR such that base pointers are available for every live pointer, we potentially need to duplicate instructions to propagate the base. The original code had only handled PHI and Select under the belief those were the only instructions which would need duplicated. When I added support for vector instructions, I'd added a collection of hacks for ExtractElement which caught most of the common cases. Of course, I then found the one test case my hacks couldn't cover. :) This change removes all of the early hacks for extract element. By defining extractelement as a BDV (rather than trying to look through it), we can extend the rewriting algorithm to duplicate the extract as needed. Note that a couple of peephole optimizations were left in for the moment, because while we now handle extractelement as a first class citizen, we're not yet handling insertelement. That change will follow in the near future. llvm-svn: 244808
2015-08-13 05:00:20 +08:00
; CHECK-LABEL: test4
; CHECK: statepoint
; CHECK: gc.relocate
[RS4GC] Revert optimization attempt due to memory corruption This change reverts "246133 [RewriteStatepointsForGC] Reduce the number of new instructions for base pointers" and a follow on bugfix 12575. As pointed out in pr25846, this code suffers from a memory corruption bug. Since I'm (empirically) not going to get back to this any time soon, simply reverting the problematic change is the right answer. llvm-svn: 261565
2016-02-23 04:45:56 +08:00
; CHECK-DAG: ; (%obj.base, %obj)
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
; When we can optimize an extractelement from a known
; index and avoid introducing new base pointer instructions
entry:
%derived = getelementptr i64, i64 addrspace(1)* %ptr, i64 16
%veca = insertelement <2 x i64 addrspace(1)*> undef, i64 addrspace(1)* %derived, i32 0
%vec = insertelement <2 x i64 addrspace(1)*> %veca, i64 addrspace(1)* %ptr, i32 1
%obj = extractelement <2 x i64 addrspace(1)*> %vec, i32 0
call void @do_safepoint() [ "deopt"() ]
[RewriteStatepointsForGC] Handle extractelement fully in the base pointer algorithm When rewriting the IR such that base pointers are available for every live pointer, we potentially need to duplicate instructions to propagate the base. The original code had only handled PHI and Select under the belief those were the only instructions which would need duplicated. When I added support for vector instructions, I'd added a collection of hacks for ExtractElement which caught most of the common cases. Of course, I then found the one test case my hacks couldn't cover. :) This change removes all of the early hacks for extract element. By defining extractelement as a BDV (rather than trying to look through it), we can extend the rewriting algorithm to duplicate the extract as needed. Note that a couple of peephole optimizations were left in for the moment, because while we now handle extractelement as a first class citizen, we're not yet handling insertelement. That change will follow in the near future. llvm-svn: 244808
2015-08-13 05:00:20 +08:00
ret i64 addrspace(1)* %obj
}
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
declare void @use(i64 addrspace(1)*) "gc-leaf-function"
[RewriteStatepointsForGC] Extend base pointer inference to handle insertelement This change is simply enhancing the existing inference algorithm to handle insertelement instructions by conservatively inserting a new instruction to propagate the vector of associated base pointers. In the process, I'm ripping out the peephole optimizations which mostly helped cover the fact this hadn't been done. Note that most of the newly inserted nodes will be nearly immediately removed by the post insertion optimization pass introduced in 246718. Arguably, we should be trying harder to avoid the malloc traffic here, but I'd rather get the code correct, then worry about compile time. Unlike previous extensions of the algorithm to handle more case, I discovered the existing code was causing miscompiles in some cases. In particular, we had an implicit assumption that the peephole covered *all* insert element instructions, so if we had a value directly based on a insert element the peephole didn't cover, we proceeded as if it were a base anyways. Not good. I believe we had the same issue with shufflevector which is why I adjusted the predicate for them as well. Differential Revision: http://reviews.llvm.org/D12583 llvm-svn: 247210
2015-09-10 07:40:12 +08:00
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
define void @test5(i1 %cnd, i64 addrspace(1)* %obj) gc "statepoint-example" {
[RewriteStatepointsForGC] Extend base pointer inference to handle insertelement This change is simply enhancing the existing inference algorithm to handle insertelement instructions by conservatively inserting a new instruction to propagate the vector of associated base pointers. In the process, I'm ripping out the peephole optimizations which mostly helped cover the fact this hadn't been done. Note that most of the newly inserted nodes will be nearly immediately removed by the post insertion optimization pass introduced in 246718. Arguably, we should be trying harder to avoid the malloc traffic here, but I'd rather get the code correct, then worry about compile time. Unlike previous extensions of the algorithm to handle more case, I discovered the existing code was causing miscompiles in some cases. In particular, we had an implicit assumption that the peephole covered *all* insert element instructions, so if we had a value directly based on a insert element the peephole didn't cover, we proceeded as if it were a base anyways. Not good. I believe we had the same issue with shufflevector which is why I adjusted the predicate for them as well. Differential Revision: http://reviews.llvm.org/D12583 llvm-svn: 247210
2015-09-10 07:40:12 +08:00
; CHECK-LABEL: @test5
; CHECK: gc.relocate
[RS4GC] Revert optimization attempt due to memory corruption This change reverts "246133 [RewriteStatepointsForGC] Reduce the number of new instructions for base pointers" and a follow on bugfix 12575. As pointed out in pr25846, this code suffers from a memory corruption bug. Since I'm (empirically) not going to get back to this any time soon, simply reverting the problematic change is the right answer. llvm-svn: 261565
2016-02-23 04:45:56 +08:00
; CHECK-DAG: (%bdv.base, %bdv)
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
; When we fundementally have to duplicate
[RewriteStatepointsForGC] Extend base pointer inference to handle insertelement This change is simply enhancing the existing inference algorithm to handle insertelement instructions by conservatively inserting a new instruction to propagate the vector of associated base pointers. In the process, I'm ripping out the peephole optimizations which mostly helped cover the fact this hadn't been done. Note that most of the newly inserted nodes will be nearly immediately removed by the post insertion optimization pass introduced in 246718. Arguably, we should be trying harder to avoid the malloc traffic here, but I'd rather get the code correct, then worry about compile time. Unlike previous extensions of the algorithm to handle more case, I discovered the existing code was causing miscompiles in some cases. In particular, we had an implicit assumption that the peephole covered *all* insert element instructions, so if we had a value directly based on a insert element the peephole didn't cover, we proceeded as if it were a base anyways. Not good. I believe we had the same issue with shufflevector which is why I adjusted the predicate for them as well. Differential Revision: http://reviews.llvm.org/D12583 llvm-svn: 247210
2015-09-10 07:40:12 +08:00
entry:
%gep = getelementptr i64, i64 addrspace(1)* %obj, i64 1
%vec = insertelement <2 x i64 addrspace(1)*> undef, i64 addrspace(1)* %gep, i32 0
%bdv = extractelement <2 x i64 addrspace(1)*> %vec, i32 0
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
call void @do_safepoint() [ "deopt"(i32 0, i32 -1, i32 0, i32 0, i32 0) ]
[RewriteStatepointsForGC] Extend base pointer inference to handle insertelement This change is simply enhancing the existing inference algorithm to handle insertelement instructions by conservatively inserting a new instruction to propagate the vector of associated base pointers. In the process, I'm ripping out the peephole optimizations which mostly helped cover the fact this hadn't been done. Note that most of the newly inserted nodes will be nearly immediately removed by the post insertion optimization pass introduced in 246718. Arguably, we should be trying harder to avoid the malloc traffic here, but I'd rather get the code correct, then worry about compile time. Unlike previous extensions of the algorithm to handle more case, I discovered the existing code was causing miscompiles in some cases. In particular, we had an implicit assumption that the peephole covered *all* insert element instructions, so if we had a value directly based on a insert element the peephole didn't cover, we proceeded as if it were a base anyways. Not good. I believe we had the same issue with shufflevector which is why I adjusted the predicate for them as well. Differential Revision: http://reviews.llvm.org/D12583 llvm-svn: 247210
2015-09-10 07:40:12 +08:00
call void @use(i64 addrspace(1)* %bdv)
ret void
}
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
define void @test6(i1 %cnd, i64 addrspace(1)* %obj, i64 %idx) gc "statepoint-example" {
[RewriteStatepointsForGC] Extend base pointer inference to handle insertelement This change is simply enhancing the existing inference algorithm to handle insertelement instructions by conservatively inserting a new instruction to propagate the vector of associated base pointers. In the process, I'm ripping out the peephole optimizations which mostly helped cover the fact this hadn't been done. Note that most of the newly inserted nodes will be nearly immediately removed by the post insertion optimization pass introduced in 246718. Arguably, we should be trying harder to avoid the malloc traffic here, but I'd rather get the code correct, then worry about compile time. Unlike previous extensions of the algorithm to handle more case, I discovered the existing code was causing miscompiles in some cases. In particular, we had an implicit assumption that the peephole covered *all* insert element instructions, so if we had a value directly based on a insert element the peephole didn't cover, we proceeded as if it were a base anyways. Not good. I believe we had the same issue with shufflevector which is why I adjusted the predicate for them as well. Differential Revision: http://reviews.llvm.org/D12583 llvm-svn: 247210
2015-09-10 07:40:12 +08:00
; CHECK-LABEL: @test6
; CHECK: %gep = getelementptr i64, i64 addrspace(1)* %obj, i64 1
[RewriteStatepointsForGC] All constant should have null base pointer Currently we consider that each constant has itself as a base value. I.e "base(const) = const". This introduces couple of problems when we are trying to avoid reporting constants in statepoint live sets: 1. When querying "base( phi(const1, const2) )" we will get "phi(const1, const2)" as a base pointer. Since it's not a constant we will record it in a stack map. However on practice we don't want this to happen (constant are never relocated). 2. base( phi(const, gc ptr) ) = phi( const, base(gc ptr) ). This particular case imposes challenge on our runtime - we don't expect to see constant base pointers other than null. This problems can be avoided by treating all constant as if they were derived from null pointer base. I.e in a first case we will not include constant pointer in a stack map at all. In a second case we will get "phi(null, base(gc ptr))" as a base pointer which is a lot more convenient. Differential Revision: http://reviews.llvm.org/D20584 llvm-svn: 270993
2016-05-27 21:13:59 +08:00
; CHECK: %vec.base = insertelement <2 x i64 addrspace(1)*> zeroinitializer, i64 addrspace(1)* %obj, i32 0, !is_base_value !0
[RewriteStatepointsForGC] Extend base pointer inference to handle insertelement This change is simply enhancing the existing inference algorithm to handle insertelement instructions by conservatively inserting a new instruction to propagate the vector of associated base pointers. In the process, I'm ripping out the peephole optimizations which mostly helped cover the fact this hadn't been done. Note that most of the newly inserted nodes will be nearly immediately removed by the post insertion optimization pass introduced in 246718. Arguably, we should be trying harder to avoid the malloc traffic here, but I'd rather get the code correct, then worry about compile time. Unlike previous extensions of the algorithm to handle more case, I discovered the existing code was causing miscompiles in some cases. In particular, we had an implicit assumption that the peephole covered *all* insert element instructions, so if we had a value directly based on a insert element the peephole didn't cover, we proceeded as if it were a base anyways. Not good. I believe we had the same issue with shufflevector which is why I adjusted the predicate for them as well. Differential Revision: http://reviews.llvm.org/D12583 llvm-svn: 247210
2015-09-10 07:40:12 +08:00
; CHECK: %vec = insertelement <2 x i64 addrspace(1)*> undef, i64 addrspace(1)* %gep, i32 0
; CHECK: %bdv.base = extractelement <2 x i64 addrspace(1)*> %vec.base, i64 %idx, !is_base_value !0
; CHECK: %bdv = extractelement <2 x i64 addrspace(1)*> %vec, i64 %idx
; CHECK: gc.statepoint
; CHECK: gc.relocate
; CHECK-DAG: (%bdv.base, %bdv)
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
; A more complicated example involving vector and scalar bases.
; This is derived from a failing test case when we didn't have correct
; insertelement handling.
[RewriteStatepointsForGC] Extend base pointer inference to handle insertelement This change is simply enhancing the existing inference algorithm to handle insertelement instructions by conservatively inserting a new instruction to propagate the vector of associated base pointers. In the process, I'm ripping out the peephole optimizations which mostly helped cover the fact this hadn't been done. Note that most of the newly inserted nodes will be nearly immediately removed by the post insertion optimization pass introduced in 246718. Arguably, we should be trying harder to avoid the malloc traffic here, but I'd rather get the code correct, then worry about compile time. Unlike previous extensions of the algorithm to handle more case, I discovered the existing code was causing miscompiles in some cases. In particular, we had an implicit assumption that the peephole covered *all* insert element instructions, so if we had a value directly based on a insert element the peephole didn't cover, we proceeded as if it were a base anyways. Not good. I believe we had the same issue with shufflevector which is why I adjusted the predicate for them as well. Differential Revision: http://reviews.llvm.org/D12583 llvm-svn: 247210
2015-09-10 07:40:12 +08:00
entry:
%gep = getelementptr i64, i64 addrspace(1)* %obj, i64 1
%vec = insertelement <2 x i64 addrspace(1)*> undef, i64 addrspace(1)* %gep, i32 0
%bdv = extractelement <2 x i64 addrspace(1)*> %vec, i64 %idx
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
call void @do_safepoint() [ "deopt"(i32 0, i32 -1, i32 0, i32 0, i32 0) ]
[RewriteStatepointsForGC] Extend base pointer inference to handle insertelement This change is simply enhancing the existing inference algorithm to handle insertelement instructions by conservatively inserting a new instruction to propagate the vector of associated base pointers. In the process, I'm ripping out the peephole optimizations which mostly helped cover the fact this hadn't been done. Note that most of the newly inserted nodes will be nearly immediately removed by the post insertion optimization pass introduced in 246718. Arguably, we should be trying harder to avoid the malloc traffic here, but I'd rather get the code correct, then worry about compile time. Unlike previous extensions of the algorithm to handle more case, I discovered the existing code was causing miscompiles in some cases. In particular, we had an implicit assumption that the peephole covered *all* insert element instructions, so if we had a value directly based on a insert element the peephole didn't cover, we proceeded as if it were a base anyways. Not good. I believe we had the same issue with shufflevector which is why I adjusted the predicate for them as well. Differential Revision: http://reviews.llvm.org/D12583 llvm-svn: 247210
2015-09-10 07:40:12 +08:00
call void @use(i64 addrspace(1)* %bdv)
ret void
}
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
define i64 addrspace(1)* @test7(i1 %cnd, i64 addrspace(1)* %obj, i64 addrspace(1)* %obj2) gc "statepoint-example" {
[RewriteStatepointsForGC] Extend base pointer inference to handle insertelement This change is simply enhancing the existing inference algorithm to handle insertelement instructions by conservatively inserting a new instruction to propagate the vector of associated base pointers. In the process, I'm ripping out the peephole optimizations which mostly helped cover the fact this hadn't been done. Note that most of the newly inserted nodes will be nearly immediately removed by the post insertion optimization pass introduced in 246718. Arguably, we should be trying harder to avoid the malloc traffic here, but I'd rather get the code correct, then worry about compile time. Unlike previous extensions of the algorithm to handle more case, I discovered the existing code was causing miscompiles in some cases. In particular, we had an implicit assumption that the peephole covered *all* insert element instructions, so if we had a value directly based on a insert element the peephole didn't cover, we proceeded as if it were a base anyways. Not good. I believe we had the same issue with shufflevector which is why I adjusted the predicate for them as well. Differential Revision: http://reviews.llvm.org/D12583 llvm-svn: 247210
2015-09-10 07:40:12 +08:00
; CHECK-LABEL: @test7
entry:
%vec = insertelement <2 x i64 addrspace(1)*> undef, i64 addrspace(1)* %obj2, i32 0
br label %merge1
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
merge1: ; preds = %merge1, %entry
[RewriteStatepointsForGC] Extend base pointer inference to handle insertelement This change is simply enhancing the existing inference algorithm to handle insertelement instructions by conservatively inserting a new instruction to propagate the vector of associated base pointers. In the process, I'm ripping out the peephole optimizations which mostly helped cover the fact this hadn't been done. Note that most of the newly inserted nodes will be nearly immediately removed by the post insertion optimization pass introduced in 246718. Arguably, we should be trying harder to avoid the malloc traffic here, but I'd rather get the code correct, then worry about compile time. Unlike previous extensions of the algorithm to handle more case, I discovered the existing code was causing miscompiles in some cases. In particular, we had an implicit assumption that the peephole covered *all* insert element instructions, so if we had a value directly based on a insert element the peephole didn't cover, we proceeded as if it were a base anyways. Not good. I believe we had the same issue with shufflevector which is why I adjusted the predicate for them as well. Differential Revision: http://reviews.llvm.org/D12583 llvm-svn: 247210
2015-09-10 07:40:12 +08:00
; CHECK-LABEL: merge1:
; CHECK: vec2.base
; CHECK: vec2
; CHECK: gep
; CHECK: vec3.base
; CHECK: vec3
%vec2 = phi <2 x i64 addrspace(1)*> [ %vec, %entry ], [ %vec3, %merge1 ]
%gep = getelementptr i64, i64 addrspace(1)* %obj2, i64 1
%vec3 = insertelement <2 x i64 addrspace(1)*> undef, i64 addrspace(1)* %gep, i32 0
br i1 %cnd, label %merge1, label %next1
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
next1: ; preds = %merge1
[RewriteStatepointsForGC] Extend base pointer inference to handle insertelement This change is simply enhancing the existing inference algorithm to handle insertelement instructions by conservatively inserting a new instruction to propagate the vector of associated base pointers. In the process, I'm ripping out the peephole optimizations which mostly helped cover the fact this hadn't been done. Note that most of the newly inserted nodes will be nearly immediately removed by the post insertion optimization pass introduced in 246718. Arguably, we should be trying harder to avoid the malloc traffic here, but I'd rather get the code correct, then worry about compile time. Unlike previous extensions of the algorithm to handle more case, I discovered the existing code was causing miscompiles in some cases. In particular, we had an implicit assumption that the peephole covered *all* insert element instructions, so if we had a value directly based on a insert element the peephole didn't cover, we proceeded as if it were a base anyways. Not good. I believe we had the same issue with shufflevector which is why I adjusted the predicate for them as well. Differential Revision: http://reviews.llvm.org/D12583 llvm-svn: 247210
2015-09-10 07:40:12 +08:00
; CHECK-LABEL: next1:
; CHECK: bdv.base =
; CHECK: bdv =
%bdv = extractelement <2 x i64 addrspace(1)*> %vec2, i32 0
br label %merge
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
merge: ; preds = %merge, %next1
[RewriteStatepointsForGC] Extend base pointer inference to handle insertelement This change is simply enhancing the existing inference algorithm to handle insertelement instructions by conservatively inserting a new instruction to propagate the vector of associated base pointers. In the process, I'm ripping out the peephole optimizations which mostly helped cover the fact this hadn't been done. Note that most of the newly inserted nodes will be nearly immediately removed by the post insertion optimization pass introduced in 246718. Arguably, we should be trying harder to avoid the malloc traffic here, but I'd rather get the code correct, then worry about compile time. Unlike previous extensions of the algorithm to handle more case, I discovered the existing code was causing miscompiles in some cases. In particular, we had an implicit assumption that the peephole covered *all* insert element instructions, so if we had a value directly based on a insert element the peephole didn't cover, we proceeded as if it were a base anyways. Not good. I believe we had the same issue with shufflevector which is why I adjusted the predicate for them as well. Differential Revision: http://reviews.llvm.org/D12583 llvm-svn: 247210
2015-09-10 07:40:12 +08:00
; CHECK-LABEL: merge:
; CHECK: %objb.base
; CHECK: %objb
; CHECK: gc.statepoint
; CHECK: gc.relocate
; CHECK-DAG: (%objb.base, %objb)
%objb = phi i64 addrspace(1)* [ %obj, %next1 ], [ %bdv, %merge ]
br i1 %cnd, label %merge, label %next
[RS4GC] Clamp UseDeoptBundles to true and update tests The full diff for the test directory may be hard to read because of the filename clash; so here's all that happened as far as the tests are concerned: ``` cd test/Transforms/RewriteStatepointsForGC git rm *ll git mv deopt-bundles/* ./ rmdir deopt-bundles find . -name '*.ll' | xargs gsed -i 's/-rs4gc-use-deopt-bundles //g' ``` llvm-svn: 259129
2016-01-29 08:28:57 +08:00
next: ; preds = %merge
call void @do_safepoint() [ "deopt"(i32 0, i32 -1, i32 0, i32 0, i32 0) ]
[RewriteStatepointsForGC] Extend base pointer inference to handle insertelement This change is simply enhancing the existing inference algorithm to handle insertelement instructions by conservatively inserting a new instruction to propagate the vector of associated base pointers. In the process, I'm ripping out the peephole optimizations which mostly helped cover the fact this hadn't been done. Note that most of the newly inserted nodes will be nearly immediately removed by the post insertion optimization pass introduced in 246718. Arguably, we should be trying harder to avoid the malloc traffic here, but I'd rather get the code correct, then worry about compile time. Unlike previous extensions of the algorithm to handle more case, I discovered the existing code was causing miscompiles in some cases. In particular, we had an implicit assumption that the peephole covered *all* insert element instructions, so if we had a value directly based on a insert element the peephole didn't cover, we proceeded as if it were a base anyways. Not good. I believe we had the same issue with shufflevector which is why I adjusted the predicate for them as well. Differential Revision: http://reviews.llvm.org/D12583 llvm-svn: 247210
2015-09-10 07:40:12 +08:00
ret i64 addrspace(1)* %objb
}
[RS4GC] Handle ShuffleVector instruction in findBasePointer Summary: This patch modifies the findBasePointer to handle the shufflevector instruction. Tests run: RS4GC tests, local downstream tests. Reviewers: reames, sanjoy Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25197 llvm-svn: 283219
2016-10-04 21:48:37 +08:00
; identify base for shufflevector
define void @test8(i64 addrspace(1)* %obj, i64 %idx) gc "statepoint-example" {
; CHECK-LABEL: @test8
; CHECK: %gep = getelementptr i64, i64 addrspace(1)* %obj, i64 1
; CHECK: %gep2 = getelementptr i64, i64 addrspace(1)* %obj, i64 2
; CHECK: %vec1.base = insertelement <4 x i64 addrspace(1)*> zeroinitializer, i64 addrspace(1)* %obj, i32 0, !is_base_value !0
; CHECK: %vec1 = insertelement <4 x i64 addrspace(1)*> undef, i64 addrspace(1)* %gep, i32 0
; CHECK: %vec2.base = insertelement <4 x i64 addrspace(1)*> zeroinitializer, i64 addrspace(1)* %obj, i32 2, !is_base_value !0
; CHECK: %vec2 = insertelement <4 x i64 addrspace(1)*> undef, i64 addrspace(1)* %gep2, i32 2
; CHECK: %vec.base = shufflevector <4 x i64 addrspace(1)*> %vec1.base, <4 x i64 addrspace(1)*> %vec2.base, <2 x i32> <i32 0, i32 2>, !is_base_value !0
; CHECK: %vec = shufflevector <4 x i64 addrspace(1)*> %vec1, <4 x i64 addrspace(1)*> %vec2, <2 x i32> <i32 0, i32 2>
; CHECK: %bdv.base = extractelement <2 x i64 addrspace(1)*> %vec.base, i64 %idx, !is_base_value !0
; CHECK: %bdv = extractelement <2 x i64 addrspace(1)*> %vec, i64 %idx
; CHECK: gc.statepoint
; CHECK: gc.relocate
; CHECK-DAG: (%bdv.base, %bdv)
entry:
%gep = getelementptr i64, i64 addrspace(1)* %obj, i64 1
%gep2 = getelementptr i64, i64 addrspace(1)* %obj, i64 2
%vec1 = insertelement <4 x i64 addrspace(1)*> undef, i64 addrspace(1)* %gep, i32 0
%vec2 = insertelement <4 x i64 addrspace(1)*> undef, i64 addrspace(1)* %gep2, i32 2
%vec = shufflevector <4 x i64 addrspace(1)*> %vec1, <4 x i64 addrspace(1)*> %vec2, <2 x i32> <i32 0, i32 2>
%bdv = extractelement <2 x i64 addrspace(1)*> %vec, i64 %idx
call void @do_safepoint() [ "deopt"(i32 0, i32 -1, i32 0, i32 0, i32 0) ]
call void @use(i64 addrspace(1)* %bdv)
ret void
}
; Since the same 'base' vector is used in the shuffle operands, we do not need
; create a shufflevector base.
define void @test9(<4 x i64 addrspace(1)*> %vec1, i64 %idx) gc "statepoint-example" {
; CHECK-LABEL: @test9
; CHECK: %vec = shufflevector <4 x i64 addrspace(1)*> %vec1, <4 x i64 addrspace(1)*> %vec1, <2 x i32> <i32 0, i32 2>
; CHECK: %base_ee = extractelement <4 x i64 addrspace(1)*> %vec1, i64 %idx, !is_base_value !0
; CHECK: %bdv = extractelement <2 x i64 addrspace(1)*> %vec, i64 %idx
; CHECK: gc.statepoint
; CHECK: gc.relocate
; CHECK-DAG: (%base_ee, %bdv)
entry:
; shrinking vec1 into vec
%vec = shufflevector <4 x i64 addrspace(1)*> %vec1, <4 x i64 addrspace(1)*> %vec1, <2 x i32> <i32 0, i32 2>
%bdv = extractelement <2 x i64 addrspace(1)*> %vec, i64 %idx
call void @do_safepoint() [ "deopt"(i32 0, i32 -1, i32 0, i32 0, i32 0) ]
call void @use(i64 addrspace(1)* %bdv)
ret void
}
; vector operand of shufflevector is a phi
define i64 addrspace(1)* @test10(i1 %cnd, i64 addrspace(1)* %obj, i64 addrspace(1)* %obj2) gc "statepoint-example" {
; CHECK-LABEL: @test10
entry:
%vec1 = insertelement <4 x i64 addrspace(1)*> undef, i64 addrspace(1)* %obj, i32 0
br i1 %cnd, label %here, label %merge
here:
%vec2 = insertelement <4 x i64 addrspace(1)*> undef, i64 addrspace(1)* %obj2, i32 2
br label %merge
merge: ; preds = %merge, %entry, %here
; CHECK-LABEL: merge:
; CHECK: %vec.base = phi <4 x i64 addrspace(1)*> [ %vec1.base, %entry ], [ %vec2.base, %here ], [ %vec3.base, %merge ], !is_base_value !0
; CHECK: vec
; CHECK: vec3.base = shufflevector <4 x i64 addrspace(1)*> %vec.base, <4 x i64 addrspace(1)*> %vec.base
; CHECK: vec3
; CHECK: bdv.base
; CHECK: bdv
%vec = phi <4 x i64 addrspace(1)*> [ %vec1, %entry ], [ %vec2, %here], [ %vec3, %merge]
%vec3 = shufflevector <4 x i64 addrspace(1)*> %vec, <4 x i64 addrspace(1)*> %vec, <4 x i32> <i32 2, i32 0, i32 1, i32 3>
%bdv = extractelement <4 x i64 addrspace(1)*> %vec3, i32 0
br i1 %cnd, label %merge, label %next
next:
; CHECK-LABEL: next:
; CHECK: gc.statepoint
; CHECK: gc.relocate
; CHECK-DAG: (%bdv.base, %bdv)
call void @do_safepoint() [ "deopt"(i32 0, i32 -1, i32 0, i32 0, i32 0) ]
ret i64 addrspace(1)* %bdv
}
[RewriteStatepointsForGC] Handle extractelement fully in the base pointer algorithm When rewriting the IR such that base pointers are available for every live pointer, we potentially need to duplicate instructions to propagate the base. The original code had only handled PHI and Select under the belief those were the only instructions which would need duplicated. When I added support for vector instructions, I'd added a collection of hacks for ExtractElement which caught most of the common cases. Of course, I then found the one test case my hacks couldn't cover. :) This change removes all of the early hacks for extract element. By defining extractelement as a BDV (rather than trying to look through it), we can extend the rewriting algorithm to duplicate the extract as needed. Note that a couple of peephole optimizations were left in for the moment, because while we now handle extractelement as a first class citizen, we're not yet handling insertelement. That change will follow in the near future. llvm-svn: 244808
2015-08-13 05:00:20 +08:00
declare void @do_safepoint()