replica
/
hanchenye-llvm-project
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
hanchenye-llvm-project/llvm/test/CodeGen/X86/fp-une-cmp.ll

136 lines
3.2 KiB
LLVM
Raw Normal View History

[x86] simplify test and tighten checks I noticed this test as part of: http://reviews.llvm.org/D11393 ...which is confusing enough as-is. Let's show the exact codegen, so the changes will be more obvious. llvm-svn: 262874
2016-03-08 06:53:23 +08:00
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s
Add testcase to make sure we don't generate too many jumps for a une compare. <rdar://problem/7859988> llvm-svn: 191040
2013-09-20 05:58:20 +08:00
; <rdar://problem/7859988>
; Make sure we don't generate more jumps than we need to. We used to generate
; something like this:
;
; jne LBB0_1
; jnp LBB0_2
; LBB0_1:
; jmp LBB0_3
; LBB0_2:
; addsd ...
; LBB0_3:
;
; Now we generate this:
;
; jne LBB0_2
; jp LBB0_2
; addsd ...
; LBB0_2:
[x86] simplify test and tighten checks I noticed this test as part of: http://reviews.llvm.org/D11393 ...which is confusing enough as-is. Let's show the exact codegen, so the changes will be more obvious. llvm-svn: 262874
2016-03-08 06:53:23 +08:00
define double @rdar_7859988(double %x, double %y) nounwind readnone optsize ssp {
; CHECK-LABEL: rdar_7859988:
; CHECK: # BB#0: # %entry
; CHECK-NEXT: mulsd %xmm1, %xmm0
; CHECK-NEXT: xorpd %xmm1, %xmm1
; CHECK-NEXT: ucomisd %xmm1, %xmm0
; CHECK-NEXT: jne .LBB0_2
; CHECK-NEXT: jp .LBB0_2
; CHECK-NEXT: # BB#1: # %bb1
; CHECK-NEXT: addsd {{.*}}(%rip), %xmm0
; CHECK-NEXT: .LBB0_2: # %bb2
; CHECK-NEXT: retq
Add testcase to make sure we don't generate too many jumps for a une compare. <rdar://problem/7859988> llvm-svn: 191040
2013-09-20 05:58:20 +08:00
Allow X86::COND_NE_OR_P and X86::COND_NP_OR_E to be reversed. Currently, AnalyzeBranch() fails non-equality comparison between floating points on X86 (see https://llvm.org/bugs/show_bug.cgi?id=23875). This is because this function can modify the branch by reversing the conditional jump and removing unconditional jump if there is a proper fall-through. However, in the case of non-equality comparison between floating points, this can turn the branch "unanalyzable". Consider the following case: jne.BB1 jp.BB1 jmp.BB2 .BB1: ... .BB2: ... AnalyzeBranch() will reverse "jp .BB1" to "jnp .BB2" and then "jmp .BB2" will be removed: jne.BB1 jnp.BB2 .BB1: ... .BB2: ... However, AnalyzeBranch() cannot analyze this branch anymore as there are two conditional jumps with different targets. This may disable some optimizations like block-placement: in this case the fall-through behavior is enforced even if the fall-through block is very cold, which is suboptimal. Actually this optimization is also done in block-placement pass, which means we can remove this optimization from AnalyzeBranch(). However, currently X86::COND_NE_OR_P and X86::COND_NP_OR_E are not reversible: there is no defined negation conditions for them. In order to reverse them, this patch defines two new CondCode X86::COND_E_AND_NP and X86::COND_P_AND_NE. It also defines how to synthesize instructions for them. Here only the second conditional jump is reversed. This is valid as we only need them to do this "unconditional jump removal" optimization. Differential Revision: http://reviews.llvm.org/D11393 llvm-svn: 258847
2016-01-27 04:08:01 +08:00
entry:
[x86] simplify test and tighten checks I noticed this test as part of: http://reviews.llvm.org/D11393 ...which is confusing enough as-is. Let's show the exact codegen, so the changes will be more obvious. llvm-svn: 262874
2016-03-08 06:53:23 +08:00
%mul = fmul double %x, %y
CodeGen: Allow small copyable blocks to "break" the CFG. When choosing the best successor for a block, ordinarily we would have preferred a block that preserves the CFG unless there is a strong probability the other direction. For small blocks that can be duplicated we now skip that requirement as well, subject to some simple frequency calculations. Differential Revision: https://reviews.llvm.org/D28583 llvm-svn: 293716
2017-02-01 07:48:32 +08:00
%cmp = fcmp oeq double %mul, 0.000000e+00
br i1 %cmp, label %bb1, label %bb2
Allow X86::COND_NE_OR_P and X86::COND_NP_OR_E to be reversed. Currently, AnalyzeBranch() fails non-equality comparison between floating points on X86 (see https://llvm.org/bugs/show_bug.cgi?id=23875). This is because this function can modify the branch by reversing the conditional jump and removing unconditional jump if there is a proper fall-through. However, in the case of non-equality comparison between floating points, this can turn the branch "unanalyzable". Consider the following case: jne.BB1 jp.BB1 jmp.BB2 .BB1: ... .BB2: ... AnalyzeBranch() will reverse "jp .BB1" to "jnp .BB2" and then "jmp .BB2" will be removed: jne.BB1 jnp.BB2 .BB1: ... .BB2: ... However, AnalyzeBranch() cannot analyze this branch anymore as there are two conditional jumps with different targets. This may disable some optimizations like block-placement: in this case the fall-through behavior is enforced even if the fall-through block is very cold, which is suboptimal. Actually this optimization is also done in block-placement pass, which means we can remove this optimization from AnalyzeBranch(). However, currently X86::COND_NE_OR_P and X86::COND_NP_OR_E are not reversible: there is no defined negation conditions for them. In order to reverse them, this patch defines two new CondCode X86::COND_E_AND_NP and X86::COND_P_AND_NE. It also defines how to synthesize instructions for them. Here only the second conditional jump is reversed. This is valid as we only need them to do this "unconditional jump removal" optimization. Differential Revision: http://reviews.llvm.org/D11393 llvm-svn: 258847
2016-01-27 04:08:01 +08:00
bb1:
[x86] simplify test and tighten checks I noticed this test as part of: http://reviews.llvm.org/D11393 ...which is confusing enough as-is. Let's show the exact codegen, so the changes will be more obvious. llvm-svn: 262874
2016-03-08 06:53:23 +08:00
%add = fadd double %mul, -1.000000e+00
Allow X86::COND_NE_OR_P and X86::COND_NP_OR_E to be reversed. Currently, AnalyzeBranch() fails non-equality comparison between floating points on X86 (see https://llvm.org/bugs/show_bug.cgi?id=23875). This is because this function can modify the branch by reversing the conditional jump and removing unconditional jump if there is a proper fall-through. However, in the case of non-equality comparison between floating points, this can turn the branch "unanalyzable". Consider the following case: jne.BB1 jp.BB1 jmp.BB2 .BB1: ... .BB2: ... AnalyzeBranch() will reverse "jp .BB1" to "jnp .BB2" and then "jmp .BB2" will be removed: jne.BB1 jnp.BB2 .BB1: ... .BB2: ... However, AnalyzeBranch() cannot analyze this branch anymore as there are two conditional jumps with different targets. This may disable some optimizations like block-placement: in this case the fall-through behavior is enforced even if the fall-through block is very cold, which is suboptimal. Actually this optimization is also done in block-placement pass, which means we can remove this optimization from AnalyzeBranch(). However, currently X86::COND_NE_OR_P and X86::COND_NP_OR_E are not reversible: there is no defined negation conditions for them. In order to reverse them, this patch defines two new CondCode X86::COND_E_AND_NP and X86::COND_P_AND_NE. It also defines how to synthesize instructions for them. Here only the second conditional jump is reversed. This is valid as we only need them to do this "unconditional jump removal" optimization. Differential Revision: http://reviews.llvm.org/D11393 llvm-svn: 258847
2016-01-27 04:08:01 +08:00
br label %bb2
bb2:
[x86] simplify test and tighten checks I noticed this test as part of: http://reviews.llvm.org/D11393 ...which is confusing enough as-is. Let's show the exact codegen, so the changes will be more obvious. llvm-svn: 262874
2016-03-08 06:53:23 +08:00
%phi = phi double [ %add, %bb1 ], [ %mul, %entry ]
ret double %phi
Allow X86::COND_NE_OR_P and X86::COND_NP_OR_E to be reversed. Currently, AnalyzeBranch() fails non-equality comparison between floating points on X86 (see https://llvm.org/bugs/show_bug.cgi?id=23875). This is because this function can modify the branch by reversing the conditional jump and removing unconditional jump if there is a proper fall-through. However, in the case of non-equality comparison between floating points, this can turn the branch "unanalyzable". Consider the following case: jne.BB1 jp.BB1 jmp.BB2 .BB1: ... .BB2: ... AnalyzeBranch() will reverse "jp .BB1" to "jnp .BB2" and then "jmp .BB2" will be removed: jne.BB1 jnp.BB2 .BB1: ... .BB2: ... However, AnalyzeBranch() cannot analyze this branch anymore as there are two conditional jumps with different targets. This may disable some optimizations like block-placement: in this case the fall-through behavior is enforced even if the fall-through block is very cold, which is suboptimal. Actually this optimization is also done in block-placement pass, which means we can remove this optimization from AnalyzeBranch(). However, currently X86::COND_NE_OR_P and X86::COND_NP_OR_E are not reversible: there is no defined negation conditions for them. In order to reverse them, this patch defines two new CondCode X86::COND_E_AND_NP and X86::COND_P_AND_NE. It also defines how to synthesize instructions for them. Here only the second conditional jump is reversed. This is valid as we only need them to do this "unconditional jump removal" optimization. Differential Revision: http://reviews.llvm.org/D11393 llvm-svn: 258847
2016-01-27 04:08:01 +08:00
}
[x86] simplify test and tighten checks I noticed this test as part of: http://reviews.llvm.org/D11393 ...which is confusing enough as-is. Let's show the exact codegen, so the changes will be more obvious. llvm-svn: 262874
2016-03-08 06:53:23 +08:00
[x86] add test to show missing optimization This should make it clearer how this proposed patch: http://reviews.llvm.org/D11393 ...will change codegen. llvm-svn: 262875
2016-03-08 07:13:06 +08:00
define double @profile_metadata(double %x, double %y) {
; CHECK-LABEL: profile_metadata:
; CHECK: # BB#0: # %entry
; CHECK-NEXT: mulsd %xmm1, %xmm0
; CHECK-NEXT: xorpd %xmm1, %xmm1
; CHECK-NEXT: ucomisd %xmm1, %xmm0
; CHECK-NEXT: jne .LBB1_1
Allow X86::COND_NE_OR_P and X86::COND_NP_OR_E to be reversed. Currently, AnalyzeBranch() fails non-equality comparison between floating points on X86 (see https://llvm.org/bugs/show_bug.cgi?id=23875). This is because this function can modify the branch by reversing the conditional jump and removing unconditional jump if there is a proper fall-through. However, in the case of non-equality comparison between floating points, this can turn the branch "unanalyzable". Consider the following case: jne.BB1 jp.BB1 jmp.BB2 .BB1: ... .BB2: ... AnalyzeBranch() will reverse "jp .BB1" to "jnp .BB2" and then "jmp .BB2" will be removed: jne.BB1 jnp.BB2 .BB1: ... .BB2: ... However, AnalyzeBranch() cannot analyze this branch anymore as there are two conditional jumps with different targets. This may disable some optimizations like block-placement: in this case the fall-through behavior is enforced even if the fall-through block is very cold, which is suboptimal. Actually this optimization is also done in block-placement pass, which means we can remove this optimization from AnalyzeBranch(). However, currently X86::COND_NE_OR_P and X86::COND_NP_OR_E are not reversible: there is no defined negation conditions for them. In order to reverse them, this patch defines two new CondCode X86::COND_E_AND_NP and X86::COND_P_AND_NE. It also defines how to synthesize instructions for them. Here only the second conditional jump is reversed. This is valid as we only need them to do this "unconditional jump removal" optimization. Differential Revision: http://reviews.llvm.org/D11393 llvm-svn: 264199
2016-03-24 05:45:37 +08:00
; CHECK-NEXT: jp .LBB1_1
Codegen: Tail-duplicate during placement. The tail duplication pass uses an assumed layout when making duplication decisions. This is fine, but passes up duplication opportunities that may arise when blocks are outlined. Because we want the updated CFG to affect subsequent placement decisions, this change must occur during placement. In order to achieve this goal, TailDuplicationPass is split into a utility class, TailDuplicator, and the pass itself. The pass delegates nearly everything to the TailDuplicator object, except for looping over the blocks in a function. This allows the same code to be used for tail duplication in both places. This change, in concert with outlining optional branches, allows triangle shaped code to perform much better, esepecially when the taken/untaken branches are correlated, as it creates a second spine when the tests are small enough. Issue from previous rollback fixed, and a new test was added for that case as well. Issue was worklist/scheduling/taildup issue in layout. Issue from 2nd rollback fixed, with 2 additional tests. Issue was tail merging/loop info/tail-duplication causing issue with loops that share a header block. Issue with early tail-duplication of blocks that branch to a fallthrough predecessor fixed with test case: tail-dup-branch-to-fallthrough.ll Differential revision: https://reviews.llvm.org/D18226 llvm-svn: 283934
2016-10-12 04:36:43 +08:00
; CHECK-NEXT: # %bb2
[x86] add test to show missing optimization This should make it clearer how this proposed patch: http://reviews.llvm.org/D11393 ...will change codegen. llvm-svn: 262875
2016-03-08 07:13:06 +08:00
; CHECK-NEXT: retq
Allow X86::COND_NE_OR_P and X86::COND_NP_OR_E to be reversed. Currently, AnalyzeBranch() fails non-equality comparison between floating points on X86 (see https://llvm.org/bugs/show_bug.cgi?id=23875). This is because this function can modify the branch by reversing the conditional jump and removing unconditional jump if there is a proper fall-through. However, in the case of non-equality comparison between floating points, this can turn the branch "unanalyzable". Consider the following case: jne.BB1 jp.BB1 jmp.BB2 .BB1: ... .BB2: ... AnalyzeBranch() will reverse "jp .BB1" to "jnp .BB2" and then "jmp .BB2" will be removed: jne.BB1 jnp.BB2 .BB1: ... .BB2: ... However, AnalyzeBranch() cannot analyze this branch anymore as there are two conditional jumps with different targets. This may disable some optimizations like block-placement: in this case the fall-through behavior is enforced even if the fall-through block is very cold, which is suboptimal. Actually this optimization is also done in block-placement pass, which means we can remove this optimization from AnalyzeBranch(). However, currently X86::COND_NE_OR_P and X86::COND_NP_OR_E are not reversible: there is no defined negation conditions for them. In order to reverse them, this patch defines two new CondCode X86::COND_E_AND_NP and X86::COND_P_AND_NE. It also defines how to synthesize instructions for them. Here only the second conditional jump is reversed. This is valid as we only need them to do this "unconditional jump removal" optimization. Differential Revision: http://reviews.llvm.org/D11393 llvm-svn: 264199
2016-03-24 05:45:37 +08:00
; CHECK-NEXT: .LBB1_1: # %bb1
; CHECK-NEXT: addsd {{.*}}(%rip), %xmm0
Codegen: Tail-duplicate during placement. The tail duplication pass uses an assumed layout when making duplication decisions. This is fine, but passes up duplication opportunities that may arise when blocks are outlined. Because we want the updated CFG to affect subsequent placement decisions, this change must occur during placement. In order to achieve this goal, TailDuplicationPass is split into a utility class, TailDuplicator, and the pass itself. The pass delegates nearly everything to the TailDuplicator object, except for looping over the blocks in a function. This allows the same code to be used for tail duplication in both places. This change, in concert with outlining optional branches, allows triangle shaped code to perform much better, esepecially when the taken/untaken branches are correlated, as it creates a second spine when the tests are small enough. Issue from previous rollback fixed, and a new test was added for that case as well. Issue was worklist/scheduling/taildup issue in layout. Issue from 2nd rollback fixed, with 2 additional tests. Issue was tail merging/loop info/tail-duplication causing issue with loops that share a header block. Issue with early tail-duplication of blocks that branch to a fallthrough predecessor fixed with test case: tail-dup-branch-to-fallthrough.ll Differential revision: https://reviews.llvm.org/D18226 llvm-svn: 283934
2016-10-12 04:36:43 +08:00
; CHECK-NEXT: retq
[x86] add test to show missing optimization This should make it clearer how this proposed patch: http://reviews.llvm.org/D11393 ...will change codegen. llvm-svn: 262875
2016-03-08 07:13:06 +08:00
entry:
%mul = fmul double %x, %y
%cmp = fcmp une double %mul, 0.000000e+00
br i1 %cmp, label %bb1, label %bb2, !prof !1
bb1:
%add = fadd double %mul, -1.000000e+00
br label %bb2
bb2:
%phi = phi double [ %add, %bb1 ], [ %mul, %entry ]
ret double %phi
}
Allow X86::COND_NE_OR_P and X86::COND_NP_OR_E to be reversed. Currently, AnalyzeBranch() fails non-equality comparison between floating points on X86 (see https://llvm.org/bugs/show_bug.cgi?id=23875). This is because this function can modify the branch by reversing the conditional jump and removing unconditional jump if there is a proper fall-through. However, in the case of non-equality comparison between floating points, this can turn the branch "unanalyzable". Consider the following case: jne.BB1 jp.BB1 jmp.BB2 .BB1: ... .BB2: ... AnalyzeBranch() will reverse "jp .BB1" to "jnp .BB2" and then "jmp .BB2" will be removed: jne.BB1 jnp.BB2 .BB1: ... .BB2: ... However, AnalyzeBranch() cannot analyze this branch anymore as there are two conditional jumps with different targets. This may disable some optimizations like block-placement: in this case the fall-through behavior is enforced even if the fall-through block is very cold, which is suboptimal. Actually this optimization is also done in block-placement pass, which means we can remove this optimization from AnalyzeBranch(). However, currently X86::COND_NE_OR_P and X86::COND_NP_OR_E are not reversible: there is no defined negation conditions for them. In order to reverse them, this patch defines two new CondCode X86::COND_E_AND_NP and X86::COND_P_AND_NE. It also defines how to synthesize instructions for them. Here only the second conditional jump is reversed. This is valid as we only need them to do this "unconditional jump removal" optimization. Differential Revision: http://reviews.llvm.org/D11393 llvm-svn: 264199
2016-03-24 05:45:37 +08:00
; Test if the negation of the non-equality check between floating points are
; translated to jnp followed by jne.
[x86] add test to show missing optimization This should make it clearer how this proposed patch: http://reviews.llvm.org/D11393 ...will change codegen. llvm-svn: 262875
2016-03-08 07:13:06 +08:00
Allow X86::COND_NE_OR_P and X86::COND_NP_OR_E to be reversed. Currently, AnalyzeBranch() fails non-equality comparison between floating points on X86 (see https://llvm.org/bugs/show_bug.cgi?id=23875). This is because this function can modify the branch by reversing the conditional jump and removing unconditional jump if there is a proper fall-through. However, in the case of non-equality comparison between floating points, this can turn the branch "unanalyzable". Consider the following case: jne.BB1 jp.BB1 jmp.BB2 .BB1: ... .BB2: ... AnalyzeBranch() will reverse "jp .BB1" to "jnp .BB2" and then "jmp .BB2" will be removed: jne.BB1 jnp.BB2 .BB1: ... .BB2: ... However, AnalyzeBranch() cannot analyze this branch anymore as there are two conditional jumps with different targets. This may disable some optimizations like block-placement: in this case the fall-through behavior is enforced even if the fall-through block is very cold, which is suboptimal. Actually this optimization is also done in block-placement pass, which means we can remove this optimization from AnalyzeBranch(). However, currently X86::COND_NE_OR_P and X86::COND_NP_OR_E are not reversible: there is no defined negation conditions for them. In order to reverse them, this patch defines two new CondCode X86::COND_E_AND_NP and X86::COND_P_AND_NE. It also defines how to synthesize instructions for them. Here only the second conditional jump is reversed. This is valid as we only need them to do this "unconditional jump removal" optimization. Differential Revision: http://reviews.llvm.org/D11393 llvm-svn: 264199
2016-03-24 05:45:37 +08:00
define void @foo(float %f) {
; CHECK-LABEL: foo:
; CHECK: # BB#0: # %entry
; CHECK-NEXT: xorps %xmm1, %xmm1
; CHECK-NEXT: ucomiss %xmm1, %xmm0
; CHECK-NEXT: jne .LBB2_2
; CHECK-NEXT: jnp .LBB2_1
; CHECK-NEXT: .LBB2_2: # %if.then
; CHECK-NEXT: jmp a # TAILCALL
; CHECK-NEXT: .LBB2_1: # %if.end
; CHECK-NEXT: retq
entry:
%cmp = fcmp une float %f, 0.000000e+00
br i1 %cmp, label %if.then, label %if.end
if.then:
tail call void @a()
br label %if.end
if.end:
ret void
}
Fix for PR27750. Correctly handle the case where the fallthrough block and target block are the same in getFallThroughMBB. Differential Revision: http://reviews.llvm.org/D20288 llvm-svn: 269760
2016-05-17 20:47:46 +08:00
; Test that an FP oeq/une conditional branch can be inverted successfully even
; when the true and false targets are the same (PR27750).
;
; CHECK-LABEL: pr27750
; CHECK: ucomiss
; CHECK-NEXT: jne [[TARGET:.*]]
; CHECK-NEXT: jp [[TARGET]]
define void @pr27750(i32* %b, float %x, i1 %y) {
entry:
br label %for.cond
for.cond:
br label %for.cond1
for.cond1:
br i1 %y, label %for.body3.lr.ph, label %for.end
for.body3.lr.ph:
store i32 0, i32* %b, align 4
br label %for.end
for.end:
; After block %for.cond gets eliminated, the two target blocks of this
; conditional block are the same.
%tobool = fcmp une float %x, 0.000000e+00
br i1 %tobool, label %for.cond, label %for.cond1
}
Allow X86::COND_NE_OR_P and X86::COND_NP_OR_E to be reversed. Currently, AnalyzeBranch() fails non-equality comparison between floating points on X86 (see https://llvm.org/bugs/show_bug.cgi?id=23875). This is because this function can modify the branch by reversing the conditional jump and removing unconditional jump if there is a proper fall-through. However, in the case of non-equality comparison between floating points, this can turn the branch "unanalyzable". Consider the following case: jne.BB1 jp.BB1 jmp.BB2 .BB1: ... .BB2: ... AnalyzeBranch() will reverse "jp .BB1" to "jnp .BB2" and then "jmp .BB2" will be removed: jne.BB1 jnp.BB2 .BB1: ... .BB2: ... However, AnalyzeBranch() cannot analyze this branch anymore as there are two conditional jumps with different targets. This may disable some optimizations like block-placement: in this case the fall-through behavior is enforced even if the fall-through block is very cold, which is suboptimal. Actually this optimization is also done in block-placement pass, which means we can remove this optimization from AnalyzeBranch(). However, currently X86::COND_NE_OR_P and X86::COND_NP_OR_E are not reversible: there is no defined negation conditions for them. In order to reverse them, this patch defines two new CondCode X86::COND_E_AND_NP and X86::COND_P_AND_NE. It also defines how to synthesize instructions for them. Here only the second conditional jump is reversed. This is valid as we only need them to do this "unconditional jump removal" optimization. Differential Revision: http://reviews.llvm.org/D11393 llvm-svn: 264199
2016-03-24 05:45:37 +08:00
declare void @a()
!1 = !{!"branch_weights", i32 1, i32 1000}