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/dagcombine-select.ll

433 lines
12 KiB
LLVM
Raw Normal View History

Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
[X86] Regenerate NOBMI/BMI combine-select tests. Test cleanup for D38128 llvm-svn: 338265
2018-07-31 00:18:38 +08:00
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -verify-machineinstrs | FileCheck %s --check-prefixes=CHECK,NOBMI -enable-var-scope
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -verify-machineinstrs -mattr=+bmi | FileCheck %s -check-prefixes=CHECK,BMI -enable-var-scope
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
define i32 @select_and1(i32 %x, i32 %y) {
; CHECK-LABEL: select_and1:
; CHECK: # %bb.0:
; CHECK-NEXT: xorl %eax, %eax
; CHECK-NEXT: cmpl $11, %edi
DAG combine "and|or (select c, -1, 0), x" -> "select c, x, 0|-1" Allowed folding for "and/or" binops with non-constant operand if arguments of select are 0/-1 values. Normally this code with "and" opcode does not get to a DAG combiner and simplified yet in the InstCombine. However AMDGPU produces it during lowering and InstCombine has no chance to optimize it out. In turn the same pattern with "or" opcode can reach DAG. Differential Revision: https://reviews.llvm.org/D48301 llvm-svn: 335250
2018-06-22 00:02:05 +08:00
; CHECK-NEXT: cmovgel %esi, %eax
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
; CHECK-NEXT: retq
%c = icmp slt i32 %x, 11
%s = select i1 %c, i32 0, i32 -1
%a = and i32 %y, %s
ret i32 %a
}
define i32 @select_and2(i32 %x, i32 %y) {
; CHECK-LABEL: select_and2:
; CHECK: # %bb.0:
; CHECK-NEXT: xorl %eax, %eax
; CHECK-NEXT: cmpl $11, %edi
DAG combine "and|or (select c, -1, 0), x" -> "select c, x, 0|-1" Allowed folding for "and/or" binops with non-constant operand if arguments of select are 0/-1 values. Normally this code with "and" opcode does not get to a DAG combiner and simplified yet in the InstCombine. However AMDGPU produces it during lowering and InstCombine has no chance to optimize it out. In turn the same pattern with "or" opcode can reach DAG. Differential Revision: https://reviews.llvm.org/D48301 llvm-svn: 335250
2018-06-22 00:02:05 +08:00
; CHECK-NEXT: cmovgel %esi, %eax
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
; CHECK-NEXT: retq
%c = icmp slt i32 %x, 11
%s = select i1 %c, i32 0, i32 -1
%a = and i32 %s, %y
ret i32 %a
}
DAG combine "and|or (select c, -1, 0), x" -> "select c, x, 0|-1" Allowed folding for "and/or" binops with non-constant operand if arguments of select are 0/-1 values. Normally this code with "and" opcode does not get to a DAG combiner and simplified yet in the InstCombine. However AMDGPU produces it during lowering and InstCombine has no chance to optimize it out. In turn the same pattern with "or" opcode can reach DAG. Differential Revision: https://reviews.llvm.org/D48301 llvm-svn: 335250
2018-06-22 00:02:05 +08:00
define i32 @select_and3(i32 %x, i32 %y) {
; CHECK-LABEL: select_and3:
; CHECK: # %bb.0:
; CHECK-NEXT: xorl %eax, %eax
; CHECK-NEXT: cmpl $11, %edi
; CHECK-NEXT: cmovll %esi, %eax
; CHECK-NEXT: retq
%c = icmp slt i32 %x, 11
%s = select i1 %c, i32 -1, i32 0
%a = and i32 %y, %s
ret i32 %a
}
define <4 x i32> @select_and_v4(i32 %x, <4 x i32> %y) {
; CHECK-LABEL: select_and_v4:
; CHECK: # %bb.0:
; CHECK-NEXT: cmpl $11, %edi
; CHECK-NEXT: xorps %xmm1, %xmm1
; CHECK-NEXT: jl .LBB3_2
; CHECK-NEXT: # %bb.1:
; CHECK-NEXT: movaps %xmm0, %xmm1
; CHECK-NEXT: .LBB3_2:
; CHECK-NEXT: movaps %xmm1, %xmm0
; CHECK-NEXT: retq
%c = icmp slt i32 %x, 11
%s = select i1 %c, <4 x i32> zeroinitializer, <4 x i32><i32 -1, i32 -1, i32 -1, i32 -1>
%a = and <4 x i32> %s, %y
ret <4 x i32> %a
}
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
define i32 @select_or1(i32 %x, i32 %y) {
; CHECK-LABEL: select_or1:
; CHECK: # %bb.0:
; CHECK-NEXT: cmpl $11, %edi
DAG combine "and|or (select c, -1, 0), x" -> "select c, x, 0|-1" Allowed folding for "and/or" binops with non-constant operand if arguments of select are 0/-1 values. Normally this code with "and" opcode does not get to a DAG combiner and simplified yet in the InstCombine. However AMDGPU produces it during lowering and InstCombine has no chance to optimize it out. In turn the same pattern with "or" opcode can reach DAG. Differential Revision: https://reviews.llvm.org/D48301 llvm-svn: 335250
2018-06-22 00:02:05 +08:00
; CHECK-NEXT: movl $-1, %eax
; CHECK-NEXT: cmovll %esi, %eax
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
; CHECK-NEXT: retq
%c = icmp slt i32 %x, 11
%s = select i1 %c, i32 0, i32 -1
%a = or i32 %y, %s
ret i32 %a
}
define i32 @select_or2(i32 %x, i32 %y) {
; CHECK-LABEL: select_or2:
; CHECK: # %bb.0:
; CHECK-NEXT: cmpl $11, %edi
DAG combine "and|or (select c, -1, 0), x" -> "select c, x, 0|-1" Allowed folding for "and/or" binops with non-constant operand if arguments of select are 0/-1 values. Normally this code with "and" opcode does not get to a DAG combiner and simplified yet in the InstCombine. However AMDGPU produces it during lowering and InstCombine has no chance to optimize it out. In turn the same pattern with "or" opcode can reach DAG. Differential Revision: https://reviews.llvm.org/D48301 llvm-svn: 335250
2018-06-22 00:02:05 +08:00
; CHECK-NEXT: movl $-1, %eax
; CHECK-NEXT: cmovll %esi, %eax
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
; CHECK-NEXT: retq
%c = icmp slt i32 %x, 11
%s = select i1 %c, i32 0, i32 -1
%a = or i32 %s, %y
ret i32 %a
}
DAG combine "and|or (select c, -1, 0), x" -> "select c, x, 0|-1" Allowed folding for "and/or" binops with non-constant operand if arguments of select are 0/-1 values. Normally this code with "and" opcode does not get to a DAG combiner and simplified yet in the InstCombine. However AMDGPU produces it during lowering and InstCombine has no chance to optimize it out. In turn the same pattern with "or" opcode can reach DAG. Differential Revision: https://reviews.llvm.org/D48301 llvm-svn: 335250
2018-06-22 00:02:05 +08:00
define i32 @select_or3(i32 %x, i32 %y) {
; CHECK-LABEL: select_or3:
; CHECK: # %bb.0:
; CHECK-NEXT: cmpl $11, %edi
; CHECK-NEXT: movl $-1, %eax
; CHECK-NEXT: cmovgel %esi, %eax
; CHECK-NEXT: retq
%c = icmp slt i32 %x, 11
%s = select i1 %c, i32 -1, i32 0
%a = or i32 %y, %s
ret i32 %a
}
define <4 x i32> @select_or_v4(i32 %x, <4 x i32> %y) {
; CHECK-LABEL: select_or_v4:
; CHECK: # %bb.0:
; CHECK-NEXT: cmpl $11, %edi
; CHECK-NEXT: jl .LBB7_2
; CHECK-NEXT: # %bb.1:
; CHECK-NEXT: pcmpeqd %xmm0, %xmm0
; CHECK-NEXT: .LBB7_2:
; CHECK-NEXT: retq
%c = icmp slt i32 %x, 11
%s = select i1 %c, <4 x i32> zeroinitializer, <4 x i32><i32 -1, i32 -1, i32 -1, i32 -1>
%a = or <4 x i32> %s, %y
ret <4 x i32> %a
}
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
define i32 @sel_constants_sub_constant_sel_constants(i1 %cond) {
; CHECK-LABEL: sel_constants_sub_constant_sel_constants:
; CHECK: # %bb.0:
; CHECK-NEXT: testb $1, %dil
Allow binop C1, (select cc, CF, CT) -> select folding Previously this folding was done only if select is a first operand. However, for non-commutative operations constant may go before select. Differential Revision: https://reviews.llvm.org/D48223 llvm-svn: 335167
2018-06-21 04:24:20 +08:00
; CHECK-NEXT: movl $9, %ecx
; CHECK-NEXT: movl $2, %eax
; CHECK-NEXT: cmovnel %ecx, %eax
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
; CHECK-NEXT: retq
%sel = select i1 %cond, i32 -4, i32 3
%bo = sub i32 5, %sel
ret i32 %bo
}
define i32 @sdiv_constant_sel_constants(i1 %cond) {
; CHECK-LABEL: sdiv_constant_sel_constants:
; CHECK: # %bb.0:
Allow binop C1, (select cc, CF, CT) -> select folding Previously this folding was done only if select is a first operand. However, for non-commutative operations constant may go before select. Differential Revision: https://reviews.llvm.org/D48223 llvm-svn: 335167
2018-06-21 04:24:20 +08:00
; CHECK-NEXT: notb %dil
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: andl $1, %eax
; CHECK-NEXT: leal (%rax,%rax,4), %eax
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
; CHECK-NEXT: retq
%sel = select i1 %cond, i32 121, i32 23
%bo = sdiv i32 120, %sel
ret i32 %bo
}
define i32 @udiv_constant_sel_constants(i1 %cond) {
; CHECK-LABEL: udiv_constant_sel_constants:
; CHECK: # %bb.0:
Allow binop C1, (select cc, CF, CT) -> select folding Previously this folding was done only if select is a first operand. However, for non-commutative operations constant may go before select. Differential Revision: https://reviews.llvm.org/D48223 llvm-svn: 335167
2018-06-21 04:24:20 +08:00
; CHECK-NEXT: notb %dil
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: andl $1, %eax
; CHECK-NEXT: leal (%rax,%rax,4), %eax
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
; CHECK-NEXT: retq
%sel = select i1 %cond, i32 -4, i32 23
%bo = udiv i32 120, %sel
ret i32 %bo
}
define i32 @srem_constant_sel_constants(i1 %cond) {
; CHECK-LABEL: srem_constant_sel_constants:
; CHECK: # %bb.0:
; CHECK-NEXT: testb $1, %dil
Allow binop C1, (select cc, CF, CT) -> select folding Previously this folding was done only if select is a first operand. However, for non-commutative operations constant may go before select. Differential Revision: https://reviews.llvm.org/D48223 llvm-svn: 335167
2018-06-21 04:24:20 +08:00
; CHECK-NEXT: movl $120, %ecx
; CHECK-NEXT: movl $5, %eax
; CHECK-NEXT: cmovnel %ecx, %eax
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
; CHECK-NEXT: retq
%sel = select i1 %cond, i32 121, i32 23
%bo = srem i32 120, %sel
ret i32 %bo
}
define i32 @urem_constant_sel_constants(i1 %cond) {
; CHECK-LABEL: urem_constant_sel_constants:
; CHECK: # %bb.0:
; CHECK-NEXT: testb $1, %dil
Allow binop C1, (select cc, CF, CT) -> select folding Previously this folding was done only if select is a first operand. However, for non-commutative operations constant may go before select. Differential Revision: https://reviews.llvm.org/D48223 llvm-svn: 335167
2018-06-21 04:24:20 +08:00
; CHECK-NEXT: movl $120, %ecx
; CHECK-NEXT: movl $5, %eax
; CHECK-NEXT: cmovnel %ecx, %eax
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
; CHECK-NEXT: retq
%sel = select i1 %cond, i32 -4, i32 23
%bo = urem i32 120, %sel
ret i32 %bo
}
Allow binop C1, (select cc, CF, CT) -> select folding Previously this folding was done only if select is a first operand. However, for non-commutative operations constant may go before select. Differential Revision: https://reviews.llvm.org/D48223 llvm-svn: 335167
2018-06-21 04:24:20 +08:00
define i32 @sel_constants_shl_constant(i1 %cond) {
; CHECK-LABEL: sel_constants_shl_constant:
; CHECK: # %bb.0:
; CHECK-NEXT: notb %dil
; CHECK-NEXT: movzbl %dil, %eax
; CHECK-NEXT: andl $1, %eax
; CHECK-NEXT: orl $2, %eax
; CHECK-NEXT: shll $8, %eax
; CHECK-NEXT: retq
%sel = select i1 %cond, i32 2, i32 3
%bo = shl i32 %sel, 8
ret i32 %bo
}
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
define i32 @shl_constant_sel_constants(i1 %cond) {
; CHECK-LABEL: shl_constant_sel_constants:
; CHECK: # %bb.0:
[X86] Use XOR for SUB (C, X) during isel if will help fold an immediate Summary: Same idea as D48529, but restricted to X86 and done very late to avoid any surprises where subtract might be better for DAG combining. This seems like the safest way to do this trick. And we consider doing it as a DAG combine later. Reviewers: spatel, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D48557 llvm-svn: 335575
2018-06-26 11:11:15 +08:00
; CHECK-NEXT: movl %edi, %ecx
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-NEXT: andb $1, %cl
; CHECK-NEXT: xorb $3, %cl
; CHECK-NEXT: movl $1, %eax
; CHECK-NEXT: # kill: def $cl killed $cl killed $ecx
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
; CHECK-NEXT: shll %cl, %eax
; CHECK-NEXT: retq
%sel = select i1 %cond, i32 2, i32 3
%bo = shl i32 1, %sel
ret i32 %bo
}
define i32 @lshr_constant_sel_constants(i1 %cond) {
; CHECK-LABEL: lshr_constant_sel_constants:
; CHECK: # %bb.0:
[X86] Use XOR for SUB (C, X) during isel if will help fold an immediate Summary: Same idea as D48529, but restricted to X86 and done very late to avoid any surprises where subtract might be better for DAG combining. This seems like the safest way to do this trick. And we consider doing it as a DAG combine later. Reviewers: spatel, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D48557 llvm-svn: 335575
2018-06-26 11:11:15 +08:00
; CHECK-NEXT: movl %edi, %ecx
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-NEXT: andb $1, %cl
; CHECK-NEXT: xorb $3, %cl
; CHECK-NEXT: movl $64, %eax
; CHECK-NEXT: # kill: def $cl killed $cl killed $ecx
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
; CHECK-NEXT: shrl %cl, %eax
; CHECK-NEXT: retq
%sel = select i1 %cond, i32 2, i32 3
%bo = lshr i32 64, %sel
ret i32 %bo
}
define i32 @ashr_constant_sel_constants(i1 %cond) {
; CHECK-LABEL: ashr_constant_sel_constants:
; CHECK: # %bb.0:
[X86] Use XOR for SUB (C, X) during isel if will help fold an immediate Summary: Same idea as D48529, but restricted to X86 and done very late to avoid any surprises where subtract might be better for DAG combining. This seems like the safest way to do this trick. And we consider doing it as a DAG combine later. Reviewers: spatel, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D48557 llvm-svn: 335575
2018-06-26 11:11:15 +08:00
; CHECK-NEXT: movl %edi, %ecx
[X86] Handle COPYs of physregs better (regalloc hints) Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
2018-09-20 02:59:08 +08:00
; CHECK-NEXT: andb $1, %cl
; CHECK-NEXT: xorb $3, %cl
; CHECK-NEXT: movl $128, %eax
; CHECK-NEXT: # kill: def $cl killed $cl killed $ecx
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
; CHECK-NEXT: shrl %cl, %eax
; CHECK-NEXT: retq
%sel = select i1 %cond, i32 2, i32 3
%bo = ashr i32 128, %sel
ret i32 %bo
}
define double @fsub_constant_sel_constants(i1 %cond) {
; CHECK-LABEL: fsub_constant_sel_constants:
; CHECK: # %bb.0:
; CHECK-NEXT: testb $1, %dil
DAG combine "and|or (select c, -1, 0), x" -> "select c, x, 0|-1" Allowed folding for "and/or" binops with non-constant operand if arguments of select are 0/-1 values. Normally this code with "and" opcode does not get to a DAG combiner and simplified yet in the InstCombine. However AMDGPU produces it during lowering and InstCombine has no chance to optimize it out. In turn the same pattern with "or" opcode can reach DAG. Differential Revision: https://reviews.llvm.org/D48301 llvm-svn: 335250
2018-06-22 00:02:05 +08:00
; CHECK-NEXT: jne .LBB17_1
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
; CHECK-NEXT: # %bb.2:
; CHECK-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
Allow binop C1, (select cc, CF, CT) -> select folding Previously this folding was done only if select is a first operand. However, for non-commutative operations constant may go before select. Differential Revision: https://reviews.llvm.org/D48223 llvm-svn: 335167
2018-06-21 04:24:20 +08:00
; CHECK-NEXT: retq
DAG combine "and|or (select c, -1, 0), x" -> "select c, x, 0|-1" Allowed folding for "and/or" binops with non-constant operand if arguments of select are 0/-1 values. Normally this code with "and" opcode does not get to a DAG combiner and simplified yet in the InstCombine. However AMDGPU produces it during lowering and InstCombine has no chance to optimize it out. In turn the same pattern with "or" opcode can reach DAG. Differential Revision: https://reviews.llvm.org/D48301 llvm-svn: 335250
2018-06-22 00:02:05 +08:00
; CHECK-NEXT: .LBB17_1:
Allow binop C1, (select cc, CF, CT) -> select folding Previously this folding was done only if select is a first operand. However, for non-commutative operations constant may go before select. Differential Revision: https://reviews.llvm.org/D48223 llvm-svn: 335167
2018-06-21 04:24:20 +08:00
; CHECK-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
; CHECK-NEXT: retq
%sel = select i1 %cond, double -4.0, double 23.3
%bo = fsub double 5.1, %sel
ret double %bo
}
define double @fdiv_constant_sel_constants(i1 %cond) {
; CHECK-LABEL: fdiv_constant_sel_constants:
; CHECK: # %bb.0:
; CHECK-NEXT: testb $1, %dil
DAG combine "and|or (select c, -1, 0), x" -> "select c, x, 0|-1" Allowed folding for "and/or" binops with non-constant operand if arguments of select are 0/-1 values. Normally this code with "and" opcode does not get to a DAG combiner and simplified yet in the InstCombine. However AMDGPU produces it during lowering and InstCombine has no chance to optimize it out. In turn the same pattern with "or" opcode can reach DAG. Differential Revision: https://reviews.llvm.org/D48301 llvm-svn: 335250
2018-06-22 00:02:05 +08:00
; CHECK-NEXT: jne .LBB18_1
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
; CHECK-NEXT: # %bb.2:
; CHECK-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
Allow binop C1, (select cc, CF, CT) -> select folding Previously this folding was done only if select is a first operand. However, for non-commutative operations constant may go before select. Differential Revision: https://reviews.llvm.org/D48223 llvm-svn: 335167
2018-06-21 04:24:20 +08:00
; CHECK-NEXT: retq
DAG combine "and|or (select c, -1, 0), x" -> "select c, x, 0|-1" Allowed folding for "and/or" binops with non-constant operand if arguments of select are 0/-1 values. Normally this code with "and" opcode does not get to a DAG combiner and simplified yet in the InstCombine. However AMDGPU produces it during lowering and InstCombine has no chance to optimize it out. In turn the same pattern with "or" opcode can reach DAG. Differential Revision: https://reviews.llvm.org/D48301 llvm-svn: 335250
2018-06-22 00:02:05 +08:00
; CHECK-NEXT: .LBB18_1:
Allow binop C1, (select cc, CF, CT) -> select folding Previously this folding was done only if select is a first operand. However, for non-commutative operations constant may go before select. Differential Revision: https://reviews.llvm.org/D48223 llvm-svn: 335167
2018-06-21 04:24:20 +08:00
; CHECK-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
; CHECK-NEXT: retq
%sel = select i1 %cond, double -4.0, double 23.3
%bo = fdiv double 5.1, %sel
ret double %bo
}
define double @frem_constant_sel_constants(i1 %cond) {
; CHECK-LABEL: frem_constant_sel_constants:
; CHECK: # %bb.0:
; CHECK-NEXT: testb $1, %dil
DAG combine "and|or (select c, -1, 0), x" -> "select c, x, 0|-1" Allowed folding for "and/or" binops with non-constant operand if arguments of select are 0/-1 values. Normally this code with "and" opcode does not get to a DAG combiner and simplified yet in the InstCombine. However AMDGPU produces it during lowering and InstCombine has no chance to optimize it out. In turn the same pattern with "or" opcode can reach DAG. Differential Revision: https://reviews.llvm.org/D48301 llvm-svn: 335250
2018-06-22 00:02:05 +08:00
; CHECK-NEXT: jne .LBB19_1
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
; CHECK-NEXT: # %bb.2:
; CHECK-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
Allow binop C1, (select cc, CF, CT) -> select folding Previously this folding was done only if select is a first operand. However, for non-commutative operations constant may go before select. Differential Revision: https://reviews.llvm.org/D48223 llvm-svn: 335167
2018-06-21 04:24:20 +08:00
; CHECK-NEXT: retq
DAG combine "and|or (select c, -1, 0), x" -> "select c, x, 0|-1" Allowed folding for "and/or" binops with non-constant operand if arguments of select are 0/-1 values. Normally this code with "and" opcode does not get to a DAG combiner and simplified yet in the InstCombine. However AMDGPU produces it during lowering and InstCombine has no chance to optimize it out. In turn the same pattern with "or" opcode can reach DAG. Differential Revision: https://reviews.llvm.org/D48301 llvm-svn: 335250
2018-06-22 00:02:05 +08:00
; CHECK-NEXT: .LBB19_1:
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
; CHECK-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
Allow binop C1, (select cc, CF, CT) -> select folding Previously this folding was done only if select is a first operand. However, for non-commutative operations constant may go before select. Differential Revision: https://reviews.llvm.org/D48223 llvm-svn: 335167
2018-06-21 04:24:20 +08:00
; CHECK-NEXT: retq
Tests for dag combine select (binop) -> select. NFC. Tests will be updated with https://reviews.llvm.org/D48223 llvm-svn: 334987
2018-06-19 05:49:07 +08:00
%sel = select i1 %cond, double -4.0, double 23.3
%bo = frem double 5.1, %sel
ret double %bo
}
[X86] The TEST instruction is eliminated when BSF/TZCNT is used Summary: These changes cover the PR#31399. Now the ffs(x) function is lowered to (x != 0) ? llvm.cttz(x) + 1 : 0 and it corresponds to the following llvm code: %cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true) %tobool = icmp eq i32 %v, 0 %.op = add nuw nsw i32 %cnt, 1 %add = select i1 %tobool, i32 0, i32 %.op and x86 asm code: bsfl %edi, %ecx addl $1, %ecx testl %edi, %edi movl $0, %eax cmovnel %ecx, %eax In this case the 'test' instruction can't be eliminated because the 'add' instruction modifies the EFLAGS, namely, ZF flag that is set by the 'bsf' instruction when 'x' is zero. We now produce the following code: bsfl %edi, %ecx movl $-1, %eax cmovnel %ecx, %eax addl $1, %eax Patch by Ivan Kulagin Reviewers: davide, craig.topper, spatel, RKSimon Reviewed By: craig.topper Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D48765 llvm-svn: 336768
2018-07-11 14:57:42 +08:00
declare i64 @llvm.cttz.i64(i64, i1)
define i64 @cttz_64_eq_select(i64 %v) nounwind {
[X86] Regenerate NOBMI/BMI combine-select tests. Test cleanup for D38128 llvm-svn: 338265
2018-07-31 00:18:38 +08:00
; NOBMI-LABEL: cttz_64_eq_select:
; NOBMI: # %bb.0:
; NOBMI-NEXT: bsfq %rdi, %rcx
; NOBMI-NEXT: movq $-1, %rax
; NOBMI-NEXT: cmovneq %rcx, %rax
; NOBMI-NEXT: addq $6, %rax
; NOBMI-NEXT: retq
;
[X86] The TEST instruction is eliminated when BSF/TZCNT is used Summary: These changes cover the PR#31399. Now the ffs(x) function is lowered to (x != 0) ? llvm.cttz(x) + 1 : 0 and it corresponds to the following llvm code: %cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true) %tobool = icmp eq i32 %v, 0 %.op = add nuw nsw i32 %cnt, 1 %add = select i1 %tobool, i32 0, i32 %.op and x86 asm code: bsfl %edi, %ecx addl $1, %ecx testl %edi, %edi movl $0, %eax cmovnel %ecx, %eax In this case the 'test' instruction can't be eliminated because the 'add' instruction modifies the EFLAGS, namely, ZF flag that is set by the 'bsf' instruction when 'x' is zero. We now produce the following code: bsfl %edi, %ecx movl $-1, %eax cmovnel %ecx, %eax addl $1, %eax Patch by Ivan Kulagin Reviewers: davide, craig.topper, spatel, RKSimon Reviewed By: craig.topper Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D48765 llvm-svn: 336768
2018-07-11 14:57:42 +08:00
; BMI-LABEL: cttz_64_eq_select:
; BMI: # %bb.0:
; BMI-NEXT: tzcntq %rdi, %rcx
; BMI-NEXT: movq $-1, %rax
[X86] Try fixing r336768 llvm-svn: 337043
2018-07-14 04:54:24 +08:00
; BMI-NEXT: cmovaeq %rcx, %rax
[X86] The TEST instruction is eliminated when BSF/TZCNT is used Summary: These changes cover the PR#31399. Now the ffs(x) function is lowered to (x != 0) ? llvm.cttz(x) + 1 : 0 and it corresponds to the following llvm code: %cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true) %tobool = icmp eq i32 %v, 0 %.op = add nuw nsw i32 %cnt, 1 %add = select i1 %tobool, i32 0, i32 %.op and x86 asm code: bsfl %edi, %ecx addl $1, %ecx testl %edi, %edi movl $0, %eax cmovnel %ecx, %eax In this case the 'test' instruction can't be eliminated because the 'add' instruction modifies the EFLAGS, namely, ZF flag that is set by the 'bsf' instruction when 'x' is zero. We now produce the following code: bsfl %edi, %ecx movl $-1, %eax cmovnel %ecx, %eax addl $1, %eax Patch by Ivan Kulagin Reviewers: davide, craig.topper, spatel, RKSimon Reviewed By: craig.topper Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D48765 llvm-svn: 336768
2018-07-11 14:57:42 +08:00
; BMI-NEXT: addq $6, %rax
; BMI-NEXT: retq
[X86] Regenerate NOBMI/BMI combine-select tests. Test cleanup for D38128 llvm-svn: 338265
2018-07-31 00:18:38 +08:00
[X86] The TEST instruction is eliminated when BSF/TZCNT is used Summary: These changes cover the PR#31399. Now the ffs(x) function is lowered to (x != 0) ? llvm.cttz(x) + 1 : 0 and it corresponds to the following llvm code: %cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true) %tobool = icmp eq i32 %v, 0 %.op = add nuw nsw i32 %cnt, 1 %add = select i1 %tobool, i32 0, i32 %.op and x86 asm code: bsfl %edi, %ecx addl $1, %ecx testl %edi, %edi movl $0, %eax cmovnel %ecx, %eax In this case the 'test' instruction can't be eliminated because the 'add' instruction modifies the EFLAGS, namely, ZF flag that is set by the 'bsf' instruction when 'x' is zero. We now produce the following code: bsfl %edi, %ecx movl $-1, %eax cmovnel %ecx, %eax addl $1, %eax Patch by Ivan Kulagin Reviewers: davide, craig.topper, spatel, RKSimon Reviewed By: craig.topper Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D48765 llvm-svn: 336768
2018-07-11 14:57:42 +08:00
%cnt = tail call i64 @llvm.cttz.i64(i64 %v, i1 true)
%tobool = icmp eq i64 %v, 0
%.op = add nuw nsw i64 %cnt, 6
%add = select i1 %tobool, i64 5, i64 %.op
ret i64 %add
}
define i64 @cttz_64_ne_select(i64 %v) nounwind {
[X86] Regenerate NOBMI/BMI combine-select tests. Test cleanup for D38128 llvm-svn: 338265
2018-07-31 00:18:38 +08:00
; NOBMI-LABEL: cttz_64_ne_select:
; NOBMI: # %bb.0:
; NOBMI-NEXT: bsfq %rdi, %rcx
; NOBMI-NEXT: movq $-1, %rax
; NOBMI-NEXT: cmovneq %rcx, %rax
; NOBMI-NEXT: addq $6, %rax
; NOBMI-NEXT: retq
;
[X86] The TEST instruction is eliminated when BSF/TZCNT is used Summary: These changes cover the PR#31399. Now the ffs(x) function is lowered to (x != 0) ? llvm.cttz(x) + 1 : 0 and it corresponds to the following llvm code: %cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true) %tobool = icmp eq i32 %v, 0 %.op = add nuw nsw i32 %cnt, 1 %add = select i1 %tobool, i32 0, i32 %.op and x86 asm code: bsfl %edi, %ecx addl $1, %ecx testl %edi, %edi movl $0, %eax cmovnel %ecx, %eax In this case the 'test' instruction can't be eliminated because the 'add' instruction modifies the EFLAGS, namely, ZF flag that is set by the 'bsf' instruction when 'x' is zero. We now produce the following code: bsfl %edi, %ecx movl $-1, %eax cmovnel %ecx, %eax addl $1, %eax Patch by Ivan Kulagin Reviewers: davide, craig.topper, spatel, RKSimon Reviewed By: craig.topper Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D48765 llvm-svn: 336768
2018-07-11 14:57:42 +08:00
; BMI-LABEL: cttz_64_ne_select:
; BMI: # %bb.0:
; BMI-NEXT: tzcntq %rdi, %rcx
; BMI-NEXT: movq $-1, %rax
[X86] Try fixing r336768 llvm-svn: 337043
2018-07-14 04:54:24 +08:00
; BMI-NEXT: cmovaeq %rcx, %rax
[X86] The TEST instruction is eliminated when BSF/TZCNT is used Summary: These changes cover the PR#31399. Now the ffs(x) function is lowered to (x != 0) ? llvm.cttz(x) + 1 : 0 and it corresponds to the following llvm code: %cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true) %tobool = icmp eq i32 %v, 0 %.op = add nuw nsw i32 %cnt, 1 %add = select i1 %tobool, i32 0, i32 %.op and x86 asm code: bsfl %edi, %ecx addl $1, %ecx testl %edi, %edi movl $0, %eax cmovnel %ecx, %eax In this case the 'test' instruction can't be eliminated because the 'add' instruction modifies the EFLAGS, namely, ZF flag that is set by the 'bsf' instruction when 'x' is zero. We now produce the following code: bsfl %edi, %ecx movl $-1, %eax cmovnel %ecx, %eax addl $1, %eax Patch by Ivan Kulagin Reviewers: davide, craig.topper, spatel, RKSimon Reviewed By: craig.topper Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D48765 llvm-svn: 336768
2018-07-11 14:57:42 +08:00
; BMI-NEXT: addq $6, %rax
; BMI-NEXT: retq
[X86] Regenerate NOBMI/BMI combine-select tests. Test cleanup for D38128 llvm-svn: 338265
2018-07-31 00:18:38 +08:00
[X86] The TEST instruction is eliminated when BSF/TZCNT is used Summary: These changes cover the PR#31399. Now the ffs(x) function is lowered to (x != 0) ? llvm.cttz(x) + 1 : 0 and it corresponds to the following llvm code: %cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true) %tobool = icmp eq i32 %v, 0 %.op = add nuw nsw i32 %cnt, 1 %add = select i1 %tobool, i32 0, i32 %.op and x86 asm code: bsfl %edi, %ecx addl $1, %ecx testl %edi, %edi movl $0, %eax cmovnel %ecx, %eax In this case the 'test' instruction can't be eliminated because the 'add' instruction modifies the EFLAGS, namely, ZF flag that is set by the 'bsf' instruction when 'x' is zero. We now produce the following code: bsfl %edi, %ecx movl $-1, %eax cmovnel %ecx, %eax addl $1, %eax Patch by Ivan Kulagin Reviewers: davide, craig.topper, spatel, RKSimon Reviewed By: craig.topper Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D48765 llvm-svn: 336768
2018-07-11 14:57:42 +08:00
%cnt = tail call i64 @llvm.cttz.i64(i64 %v, i1 true)
%tobool = icmp ne i64 %v, 0
%.op = add nuw nsw i64 %cnt, 6
%add = select i1 %tobool, i64 %.op, i64 5
ret i64 %add
}
declare i32 @llvm.cttz.i32(i32, i1)
define i32 @cttz_32_eq_select(i32 %v) nounwind {
[X86] Regenerate NOBMI/BMI combine-select tests. Test cleanup for D38128 llvm-svn: 338265
2018-07-31 00:18:38 +08:00
; NOBMI-LABEL: cttz_32_eq_select:
; NOBMI: # %bb.0:
; NOBMI-NEXT: bsfl %edi, %ecx
; NOBMI-NEXT: movl $-1, %eax
; NOBMI-NEXT: cmovnel %ecx, %eax
; NOBMI-NEXT: addl $6, %eax
; NOBMI-NEXT: retq
;
[X86] The TEST instruction is eliminated when BSF/TZCNT is used Summary: These changes cover the PR#31399. Now the ffs(x) function is lowered to (x != 0) ? llvm.cttz(x) + 1 : 0 and it corresponds to the following llvm code: %cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true) %tobool = icmp eq i32 %v, 0 %.op = add nuw nsw i32 %cnt, 1 %add = select i1 %tobool, i32 0, i32 %.op and x86 asm code: bsfl %edi, %ecx addl $1, %ecx testl %edi, %edi movl $0, %eax cmovnel %ecx, %eax In this case the 'test' instruction can't be eliminated because the 'add' instruction modifies the EFLAGS, namely, ZF flag that is set by the 'bsf' instruction when 'x' is zero. We now produce the following code: bsfl %edi, %ecx movl $-1, %eax cmovnel %ecx, %eax addl $1, %eax Patch by Ivan Kulagin Reviewers: davide, craig.topper, spatel, RKSimon Reviewed By: craig.topper Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D48765 llvm-svn: 336768
2018-07-11 14:57:42 +08:00
; BMI-LABEL: cttz_32_eq_select:
; BMI: # %bb.0:
; BMI-NEXT: tzcntl %edi, %ecx
; BMI-NEXT: movl $-1, %eax
[X86] Try fixing r336768 llvm-svn: 337043
2018-07-14 04:54:24 +08:00
; BMI-NEXT: cmovael %ecx, %eax
[X86] The TEST instruction is eliminated when BSF/TZCNT is used Summary: These changes cover the PR#31399. Now the ffs(x) function is lowered to (x != 0) ? llvm.cttz(x) + 1 : 0 and it corresponds to the following llvm code: %cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true) %tobool = icmp eq i32 %v, 0 %.op = add nuw nsw i32 %cnt, 1 %add = select i1 %tobool, i32 0, i32 %.op and x86 asm code: bsfl %edi, %ecx addl $1, %ecx testl %edi, %edi movl $0, %eax cmovnel %ecx, %eax In this case the 'test' instruction can't be eliminated because the 'add' instruction modifies the EFLAGS, namely, ZF flag that is set by the 'bsf' instruction when 'x' is zero. We now produce the following code: bsfl %edi, %ecx movl $-1, %eax cmovnel %ecx, %eax addl $1, %eax Patch by Ivan Kulagin Reviewers: davide, craig.topper, spatel, RKSimon Reviewed By: craig.topper Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D48765 llvm-svn: 336768
2018-07-11 14:57:42 +08:00
; BMI-NEXT: addl $6, %eax
; BMI-NEXT: retq
[X86] Regenerate NOBMI/BMI combine-select tests. Test cleanup for D38128 llvm-svn: 338265
2018-07-31 00:18:38 +08:00
[X86] The TEST instruction is eliminated when BSF/TZCNT is used Summary: These changes cover the PR#31399. Now the ffs(x) function is lowered to (x != 0) ? llvm.cttz(x) + 1 : 0 and it corresponds to the following llvm code: %cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true) %tobool = icmp eq i32 %v, 0 %.op = add nuw nsw i32 %cnt, 1 %add = select i1 %tobool, i32 0, i32 %.op and x86 asm code: bsfl %edi, %ecx addl $1, %ecx testl %edi, %edi movl $0, %eax cmovnel %ecx, %eax In this case the 'test' instruction can't be eliminated because the 'add' instruction modifies the EFLAGS, namely, ZF flag that is set by the 'bsf' instruction when 'x' is zero. We now produce the following code: bsfl %edi, %ecx movl $-1, %eax cmovnel %ecx, %eax addl $1, %eax Patch by Ivan Kulagin Reviewers: davide, craig.topper, spatel, RKSimon Reviewed By: craig.topper Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D48765 llvm-svn: 336768
2018-07-11 14:57:42 +08:00
%cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true)
%tobool = icmp eq i32 %v, 0
%.op = add nuw nsw i32 %cnt, 6
%add = select i1 %tobool, i32 5, i32 %.op
ret i32 %add
}
define i32 @cttz_32_ne_select(i32 %v) nounwind {
[X86] Regenerate NOBMI/BMI combine-select tests. Test cleanup for D38128 llvm-svn: 338265
2018-07-31 00:18:38 +08:00
; NOBMI-LABEL: cttz_32_ne_select:
; NOBMI: # %bb.0:
; NOBMI-NEXT: bsfl %edi, %ecx
; NOBMI-NEXT: movl $-1, %eax
; NOBMI-NEXT: cmovnel %ecx, %eax
; NOBMI-NEXT: addl $6, %eax
; NOBMI-NEXT: retq
;
[X86] The TEST instruction is eliminated when BSF/TZCNT is used Summary: These changes cover the PR#31399. Now the ffs(x) function is lowered to (x != 0) ? llvm.cttz(x) + 1 : 0 and it corresponds to the following llvm code: %cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true) %tobool = icmp eq i32 %v, 0 %.op = add nuw nsw i32 %cnt, 1 %add = select i1 %tobool, i32 0, i32 %.op and x86 asm code: bsfl %edi, %ecx addl $1, %ecx testl %edi, %edi movl $0, %eax cmovnel %ecx, %eax In this case the 'test' instruction can't be eliminated because the 'add' instruction modifies the EFLAGS, namely, ZF flag that is set by the 'bsf' instruction when 'x' is zero. We now produce the following code: bsfl %edi, %ecx movl $-1, %eax cmovnel %ecx, %eax addl $1, %eax Patch by Ivan Kulagin Reviewers: davide, craig.topper, spatel, RKSimon Reviewed By: craig.topper Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D48765 llvm-svn: 336768
2018-07-11 14:57:42 +08:00
; BMI-LABEL: cttz_32_ne_select:
; BMI: # %bb.0:
; BMI-NEXT: tzcntl %edi, %ecx
; BMI-NEXT: movl $-1, %eax
[X86] Try fixing r336768 llvm-svn: 337043
2018-07-14 04:54:24 +08:00
; BMI-NEXT: cmovael %ecx, %eax
[X86] The TEST instruction is eliminated when BSF/TZCNT is used Summary: These changes cover the PR#31399. Now the ffs(x) function is lowered to (x != 0) ? llvm.cttz(x) + 1 : 0 and it corresponds to the following llvm code: %cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true) %tobool = icmp eq i32 %v, 0 %.op = add nuw nsw i32 %cnt, 1 %add = select i1 %tobool, i32 0, i32 %.op and x86 asm code: bsfl %edi, %ecx addl $1, %ecx testl %edi, %edi movl $0, %eax cmovnel %ecx, %eax In this case the 'test' instruction can't be eliminated because the 'add' instruction modifies the EFLAGS, namely, ZF flag that is set by the 'bsf' instruction when 'x' is zero. We now produce the following code: bsfl %edi, %ecx movl $-1, %eax cmovnel %ecx, %eax addl $1, %eax Patch by Ivan Kulagin Reviewers: davide, craig.topper, spatel, RKSimon Reviewed By: craig.topper Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D48765 llvm-svn: 336768
2018-07-11 14:57:42 +08:00
; BMI-NEXT: addl $6, %eax
; BMI-NEXT: retq
[X86] Regenerate NOBMI/BMI combine-select tests. Test cleanup for D38128 llvm-svn: 338265
2018-07-31 00:18:38 +08:00
[X86] The TEST instruction is eliminated when BSF/TZCNT is used Summary: These changes cover the PR#31399. Now the ffs(x) function is lowered to (x != 0) ? llvm.cttz(x) + 1 : 0 and it corresponds to the following llvm code: %cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true) %tobool = icmp eq i32 %v, 0 %.op = add nuw nsw i32 %cnt, 1 %add = select i1 %tobool, i32 0, i32 %.op and x86 asm code: bsfl %edi, %ecx addl $1, %ecx testl %edi, %edi movl $0, %eax cmovnel %ecx, %eax In this case the 'test' instruction can't be eliminated because the 'add' instruction modifies the EFLAGS, namely, ZF flag that is set by the 'bsf' instruction when 'x' is zero. We now produce the following code: bsfl %edi, %ecx movl $-1, %eax cmovnel %ecx, %eax addl $1, %eax Patch by Ivan Kulagin Reviewers: davide, craig.topper, spatel, RKSimon Reviewed By: craig.topper Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D48765 llvm-svn: 336768
2018-07-11 14:57:42 +08:00
%cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true)
%tobool = icmp ne i32 %v, 0
%.op = add nuw nsw i32 %cnt, 6
%add = select i1 %tobool, i32 %.op, i32 5
ret i32 %add
}
[X86] Add test cases for the patterns used by __builtin_ffs. We previously had tests for "__builtin_ffs + 5", but the SelectinoDAG without an extra addition came out slightly different. llvm-svn: 338612
2018-08-02 02:38:43 +08:00
; This matches the pattern emitted for __builtin_ffs
define i32 @cttz_32_eq_select_ffs(i32 %v) nounwind {
; NOBMI-LABEL: cttz_32_eq_select_ffs:
; NOBMI: # %bb.0:
[X86] Canonicalize the pattern for __builtin_ffs in a similar way to '__builtin_ffs + 5' We now emit a move of -1 before the cmov and do the addition after the cmov just like the case with an extra addition. This may be slightly worse for code size, but is more consistent with other compilers. And we might be able to hoist the mov -1 outside of loops. llvm-svn: 338613
2018-08-02 02:38:46 +08:00
; NOBMI-NEXT: bsfl %edi, %ecx
; NOBMI-NEXT: movl $-1, %eax
; NOBMI-NEXT: cmovnel %ecx, %eax
[X86] Add test cases for the patterns used by __builtin_ffs. We previously had tests for "__builtin_ffs + 5", but the SelectinoDAG without an extra addition came out slightly different. llvm-svn: 338612
2018-08-02 02:38:43 +08:00
; NOBMI-NEXT: incl %eax
; NOBMI-NEXT: retq
;
; BMI-LABEL: cttz_32_eq_select_ffs:
; BMI: # %bb.0:
[X86] Canonicalize the pattern for __builtin_ffs in a similar way to '__builtin_ffs + 5' We now emit a move of -1 before the cmov and do the addition after the cmov just like the case with an extra addition. This may be slightly worse for code size, but is more consistent with other compilers. And we might be able to hoist the mov -1 outside of loops. llvm-svn: 338613
2018-08-02 02:38:46 +08:00
; BMI-NEXT: tzcntl %edi, %ecx
; BMI-NEXT: movl $-1, %eax
; BMI-NEXT: cmovael %ecx, %eax
[X86] Add test cases for the patterns used by __builtin_ffs. We previously had tests for "__builtin_ffs + 5", but the SelectinoDAG without an extra addition came out slightly different. llvm-svn: 338612
2018-08-02 02:38:43 +08:00
; BMI-NEXT: incl %eax
; BMI-NEXT: retq
%cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true)
%tobool = icmp eq i32 %v, 0
%.op = add nuw nsw i32 %cnt, 1
%add = select i1 %tobool, i32 0, i32 %.op
ret i32 %add
}
define i32 @cttz_32_ne_select_ffs(i32 %v) nounwind {
; NOBMI-LABEL: cttz_32_ne_select_ffs:
; NOBMI: # %bb.0:
[X86] Canonicalize the pattern for __builtin_ffs in a similar way to '__builtin_ffs + 5' We now emit a move of -1 before the cmov and do the addition after the cmov just like the case with an extra addition. This may be slightly worse for code size, but is more consistent with other compilers. And we might be able to hoist the mov -1 outside of loops. llvm-svn: 338613
2018-08-02 02:38:46 +08:00
; NOBMI-NEXT: bsfl %edi, %ecx
; NOBMI-NEXT: movl $-1, %eax
; NOBMI-NEXT: cmovnel %ecx, %eax
[X86] Add test cases for the patterns used by __builtin_ffs. We previously had tests for "__builtin_ffs + 5", but the SelectinoDAG without an extra addition came out slightly different. llvm-svn: 338612
2018-08-02 02:38:43 +08:00
; NOBMI-NEXT: incl %eax
; NOBMI-NEXT: retq
;
; BMI-LABEL: cttz_32_ne_select_ffs:
; BMI: # %bb.0:
[X86] Canonicalize the pattern for __builtin_ffs in a similar way to '__builtin_ffs + 5' We now emit a move of -1 before the cmov and do the addition after the cmov just like the case with an extra addition. This may be slightly worse for code size, but is more consistent with other compilers. And we might be able to hoist the mov -1 outside of loops. llvm-svn: 338613
2018-08-02 02:38:46 +08:00
; BMI-NEXT: tzcntl %edi, %ecx
; BMI-NEXT: movl $-1, %eax
; BMI-NEXT: cmovael %ecx, %eax
[X86] Add test cases for the patterns used by __builtin_ffs. We previously had tests for "__builtin_ffs + 5", but the SelectinoDAG without an extra addition came out slightly different. llvm-svn: 338612
2018-08-02 02:38:43 +08:00
; BMI-NEXT: incl %eax
; BMI-NEXT: retq
%cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true)
%tobool = icmp ne i32 %v, 0
%.op = add nuw nsw i32 %cnt, 1
%add = select i1 %tobool, i32 %.op, i32 0
ret i32 %add
}