llvm-project/llvm/test/CodeGen/AArch64/arm64-fast-isel-call.ll

270 lines
8.2 KiB
LLVM
Raw Normal View History

; RUN: llc -fast-isel-sink-local-values -O0 -fast-isel -fast-isel-abort=2 -code-model=small -verify-machineinstrs -frame-pointer=all -mtriple=arm64-apple-darwin < %s | FileCheck %s
; RUN: llc -fast-isel-sink-local-values -O0 -fast-isel -fast-isel-abort=2 -code-model=large -verify-machineinstrs -frame-pointer=all -mtriple=arm64-apple-darwin < %s | FileCheck %s --check-prefix=LARGE
; RUN: llc -fast-isel-sink-local-values -O0 -fast-isel -fast-isel-abort=2 -code-model=small -verify-machineinstrs -frame-pointer=all -mtriple=aarch64_be-linux-gnu < %s | FileCheck %s --check-prefix=CHECK-BE
define void @call0() nounwind {
entry:
ret void
}
define void @foo0() nounwind {
entry:
; CHECK-LABEL: foo0
; CHECK: bl _call0
; LARGE-LABEL: foo0
; LARGE: adrp [[REG0:x[0-9]+]], _call0@GOTPAGE
; LARGE: ldr [[REG1:x[0-9]+]], {{\[}}[[REG0]], _call0@GOTPAGEOFF{{\]}}
; LARGE-NEXT: blr [[REG1]]
call void @call0()
ret void
}
define i32 @call1(i32 %a) nounwind {
entry:
%a.addr = alloca i32, align 4
store i32 %a, i32* %a.addr, align 4
%tmp = load i32, i32* %a.addr, align 4
ret i32 %tmp
}
define i32 @foo1(i32 %a) nounwind {
entry:
; CHECK-LABEL: foo1
; CHECK: stur w0, [x29, #-4]
; CHECK-NEXT: ldur w0, [x29, #-4]
; CHECK-NEXT: bl _call1
%a.addr = alloca i32, align 4
store i32 %a, i32* %a.addr, align 4
%tmp = load i32, i32* %a.addr, align 4
%call = call i32 @call1(i32 %tmp)
ret i32 %call
}
define i32 @sext_(i8 %a, i16 %b) nounwind {
entry:
; CHECK-LABEL: sext_
; CHECK: sxtb w0, w0
; CHECK: sxth w1, w1
; CHECK: bl _foo_sext_
call void @foo_sext_(i8 signext %a, i16 signext %b)
ret i32 0
}
declare void @foo_sext_(i8 %a, i16 %b)
define i32 @zext_(i8 %a, i16 %b) nounwind {
entry:
; CHECK-LABEL: zext_
; CHECK: uxtb w0, w0
; CHECK: uxth w1, w1
call void @foo_zext_(i8 zeroext %a, i16 zeroext %b)
ret i32 0
}
declare void @foo_zext_(i8 %a, i16 %b)
define i32 @t1(i32 %argc, i8** nocapture %argv) {
entry:
; CHECK-LABEL: @t1
; The last parameter will be passed on stack via i8.
; CHECK: strb w{{[0-9]+}}, [sp]
; CHECK: bl _bar
%call = call i32 @bar(i8 zeroext 0, i8 zeroext -8, i8 zeroext -69, i8 zeroext 28, i8 zeroext 40, i8 zeroext -70, i8 zeroext 28, i8 zeroext 39, i8 zeroext -41)
ret i32 0
}
declare i32 @bar(i8 zeroext, i8 zeroext, i8 zeroext, i8 zeroext, i8 zeroext, i8 zeroext, i8 zeroext, i8 zeroext, i8 zeroext)
; Test materialization of integers. Target-independent selector handles this.
define i32 @t2() {
entry:
; CHECK-LABEL: t2
; CHECK: mov [[REG1:x[0-9]+]], xzr
[FastISel] Sink local value materializations to first use Summary: Local values are constants, global addresses, and stack addresses that can't be folded into the instruction that uses them. For example, when storing the address of a global variable into memory, we need to materialize that address into a register. FastISel doesn't want to materialize any given local value more than once, so it generates all local value materialization code at EmitStartPt, which always dominates the current insertion point. This allows it to maintain a map of local value registers, and it knows that the local value area will always dominate the current insertion point. The downside is that local value instructions are always emitted without a source location. This is done to prevent jumpy line tables, but it means that the local value area will be considered part of the previous statement. Consider this C code: call1(); // line 1 ++global; // line 2 ++global; // line 3 call2(&global, &local); // line 4 Today we end up with assembly and line tables like this: .loc 1 1 callq call1 leaq global(%rip), %rdi leaq local(%rsp), %rsi .loc 1 2 addq $1, global(%rip) .loc 1 3 addq $1, global(%rip) .loc 1 4 callq call2 The LEA instructions in the local value area have no source location and are treated as being on line 1. Stepping through the code in a debugger and correlating it with the assembly won't make much sense, because these materializations are only required for line 4. This is actually problematic for the VS debugger "set next statement" feature, which effectively assumes that there are no registers live across statement boundaries. By sinking the local value code into the statement and fixing up the source location, we can make that feature work. This was filed as https://bugs.llvm.org/show_bug.cgi?id=35975 and https://crbug.com/793819. This change is obviously not enough to make this feature work reliably in all cases, but I felt that it was worth doing anyway because it usually generates smaller, more comprehensible -O0 code. I measured a 0.12% regression in code generation time with LLC on the sqlite3 amalgamation, so I think this is worth doing. There are some special cases worth calling out in the commit message: 1. local values materialized for phis 2. local values used by no-op casts 3. dead local value code Local values can be materialized for phis, and this does not show up as a vreg use in MachineRegisterInfo. In this case, if there are no other uses, this patch sinks the value to the first terminator, EH label, or the end of the BB if nothing else exists. Local values may also be used by no-op casts, which adds the register to the RegFixups table. Without reversing the RegFixups map direction, we don't have enough information to sink these instructions. Lastly, if the local value register has no other uses, we can delete it. This comes up when fastisel tries two instruction selection approaches and the first materializes the value but fails and the second succeeds without using the local value. Reviewers: aprantl, dblaikie, qcolombet, MatzeB, vsk, echristo Subscribers: dotdash, chandlerc, hans, sdardis, amccarth, javed.absar, zturner, llvm-commits, hiraditya Differential Revision: https://reviews.llvm.org/D43093 llvm-svn: 327581
2018-03-15 05:54:21 +08:00
; CHECK: mov x0, [[REG1]]
; CHECK: mov w1, #-8
; CHECK: mov [[REG2:w[0-9]+]], #1023
; CHECK: uxth w2, [[REG2]]
; CHECK: mov [[REG3:w[0-9]+]], #2
; CHECK: sxtb w3, [[REG3]]
[FastISel] Sink local value materializations to first use Summary: Local values are constants, global addresses, and stack addresses that can't be folded into the instruction that uses them. For example, when storing the address of a global variable into memory, we need to materialize that address into a register. FastISel doesn't want to materialize any given local value more than once, so it generates all local value materialization code at EmitStartPt, which always dominates the current insertion point. This allows it to maintain a map of local value registers, and it knows that the local value area will always dominate the current insertion point. The downside is that local value instructions are always emitted without a source location. This is done to prevent jumpy line tables, but it means that the local value area will be considered part of the previous statement. Consider this C code: call1(); // line 1 ++global; // line 2 ++global; // line 3 call2(&global, &local); // line 4 Today we end up with assembly and line tables like this: .loc 1 1 callq call1 leaq global(%rip), %rdi leaq local(%rsp), %rsi .loc 1 2 addq $1, global(%rip) .loc 1 3 addq $1, global(%rip) .loc 1 4 callq call2 The LEA instructions in the local value area have no source location and are treated as being on line 1. Stepping through the code in a debugger and correlating it with the assembly won't make much sense, because these materializations are only required for line 4. This is actually problematic for the VS debugger "set next statement" feature, which effectively assumes that there are no registers live across statement boundaries. By sinking the local value code into the statement and fixing up the source location, we can make that feature work. This was filed as https://bugs.llvm.org/show_bug.cgi?id=35975 and https://crbug.com/793819. This change is obviously not enough to make this feature work reliably in all cases, but I felt that it was worth doing anyway because it usually generates smaller, more comprehensible -O0 code. I measured a 0.12% regression in code generation time with LLC on the sqlite3 amalgamation, so I think this is worth doing. There are some special cases worth calling out in the commit message: 1. local values materialized for phis 2. local values used by no-op casts 3. dead local value code Local values can be materialized for phis, and this does not show up as a vreg use in MachineRegisterInfo. In this case, if there are no other uses, this patch sinks the value to the first terminator, EH label, or the end of the BB if nothing else exists. Local values may also be used by no-op casts, which adds the register to the RegFixups table. Without reversing the RegFixups map direction, we don't have enough information to sink these instructions. Lastly, if the local value register has no other uses, we can delete it. This comes up when fastisel tries two instruction selection approaches and the first materializes the value but fails and the second succeeds without using the local value. Reviewers: aprantl, dblaikie, qcolombet, MatzeB, vsk, echristo Subscribers: dotdash, chandlerc, hans, sdardis, amccarth, javed.absar, zturner, llvm-commits, hiraditya Differential Revision: https://reviews.llvm.org/D43093 llvm-svn: 327581
2018-03-15 05:54:21 +08:00
; CHECK: mov [[REG4:w[0-9]+]], wzr
; CHECK: and w4, [[REG4]], #0x1
; CHECK: mov [[REG5:w[0-9]+]], #1
; CHECK: and w5, [[REG5]], #0x1
; CHECK: bl _func2
%call = call i32 @func2(i64 zeroext 0, i32 signext -8, i16 zeroext 1023, i8 signext -254, i1 zeroext 0, i1 zeroext 1)
ret i32 0
}
declare i32 @func2(i64 zeroext, i32 signext, i16 zeroext, i8 signext, i1 zeroext, i1 zeroext)
declare void @callee_b0f(i8 %bp10, i8 %bp11, i8 %bp12, i8 %bp13, i8 %bp14, i8 %bp15, i8 %bp17, i8 %bp18, i8 %bp19)
define void @caller_b1f() {
entry:
; CHECK-BE-LABEL: caller_b1f
; CHECK-BE: strb w{{.*}}, [sp, #7]
call void @callee_b0f(i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 42)
ret void
}
define zeroext i1 @call_arguments1(i1 %a1, i1 %a2, i1 %a3, i1 %a4, i1 %a5, i1 %a6, i1 %a7, i1 %a8) {
; CHECK-LABEL: call_arguments1
; CHECK: and {{w[0-9]+}}, w0, w1
; CHECK-NEXT: and {{w[0-9]+}}, w2, w3
; CHECK-NEXT: and {{w[0-9]+}}, w4, w5
; CHECK-NEXT: and {{w[0-9]+}}, w6, w7
%1 = and i1 %a1, %a2
%2 = and i1 %a3, %a4
%3 = and i1 %a5, %a6
%4 = and i1 %a7, %a8
%5 = and i1 %1, %2
%6 = and i1 %3, %4
%7 = and i1 %5, %6
ret i1 %7
}
define i32 @call_arguments2(i8 zeroext %a1, i8 zeroext %a2, i8 zeroext %a3, i8 zeroext %a4, i8 signext %a5, i8 signext %a6, i8 signext %a7, i8 signext %a8) {
; CHECK-LABEL: call_arguments2
; CHECK: add {{w[0-9]+}}, w0, w1
; CHECK-NEXT: add {{w[0-9]+}}, w2, w3
; CHECK-NEXT: add {{w[0-9]+}}, w4, w5
; CHECK-NEXT: add {{w[0-9]+}}, w6, w7
%a1z = zext i8 %a1 to i32
%a2z = zext i8 %a2 to i32
%a3z = zext i8 %a3 to i32
%a4z = zext i8 %a4 to i32
%a5s = sext i8 %a5 to i32
%a6s = sext i8 %a6 to i32
%a7s = sext i8 %a7 to i32
%a8s = sext i8 %a8 to i32
%1 = add i32 %a1z, %a2z
%2 = add i32 %a3z, %a4z
%3 = add i32 %a5s, %a6s
%4 = add i32 %a7s, %a8s
%5 = add i32 %1, %2
%6 = add i32 %3, %4
%7 = add i32 %5, %6
ret i32 %7
}
define i32 @call_arguments3(i16 zeroext %a1, i16 zeroext %a2, i16 zeroext %a3, i16 zeroext %a4, i16 signext %a5, i16 signext %a6, i16 signext %a7, i16 signext %a8) {
; CHECK-LABEL: call_arguments3
; CHECK: add {{w[0-9]+}}, w0, w1
; CHECK-NEXT: add {{w[0-9]+}}, w2, w3
; CHECK-NEXT: add {{w[0-9]+}}, w4, w5
; CHECK-NEXT: add {{w[0-9]+}}, w6, w7
%a1z = zext i16 %a1 to i32
%a2z = zext i16 %a2 to i32
%a3z = zext i16 %a3 to i32
%a4z = zext i16 %a4 to i32
%a5s = sext i16 %a5 to i32
%a6s = sext i16 %a6 to i32
%a7s = sext i16 %a7 to i32
%a8s = sext i16 %a8 to i32
%1 = add i32 %a1z, %a2z
%2 = add i32 %a3z, %a4z
%3 = add i32 %a5s, %a6s
%4 = add i32 %a7s, %a8s
%5 = add i32 %1, %2
%6 = add i32 %3, %4
%7 = add i32 %5, %6
ret i32 %7
}
define i32 @call_arguments4(i32 %a1, i32 %a2, i32 %a3, i32 %a4, i32 %a5, i32 %a6, i32 %a7, i32 %a8) {
; CHECK-LABEL: call_arguments4
; CHECK: add {{w[0-9]+}}, w0, w1
; CHECK-NEXT: add {{w[0-9]+}}, w2, w3
; CHECK-NEXT: add {{w[0-9]+}}, w4, w5
; CHECK-NEXT: add {{w[0-9]+}}, w6, w7
%1 = add i32 %a1, %a2
%2 = add i32 %a3, %a4
%3 = add i32 %a5, %a6
%4 = add i32 %a7, %a8
%5 = add i32 %1, %2
%6 = add i32 %3, %4
%7 = add i32 %5, %6
ret i32 %7
}
define i64 @call_arguments5(i64 %a1, i64 %a2, i64 %a3, i64 %a4, i64 %a5, i64 %a6, i64 %a7, i64 %a8) {
; CHECK-LABEL: call_arguments5
; CHECK: add {{x[0-9]+}}, x0, x1
; CHECK-NEXT: add {{x[0-9]+}}, x2, x3
; CHECK-NEXT: add {{x[0-9]+}}, x4, x5
; CHECK-NEXT: add {{x[0-9]+}}, x6, x7
%1 = add i64 %a1, %a2
%2 = add i64 %a3, %a4
%3 = add i64 %a5, %a6
%4 = add i64 %a7, %a8
%5 = add i64 %1, %2
%6 = add i64 %3, %4
%7 = add i64 %5, %6
ret i64 %7
}
define float @call_arguments6(float %a1, float %a2, float %a3, float %a4, float %a5, float %a6, float %a7, float %a8) {
; CHECK-LABEL: call_arguments6
; CHECK: fadd {{s[0-9]+}}, s0, s1
; CHECK-NEXT: fadd {{s[0-9]+}}, s2, s3
; CHECK-NEXT: fadd {{s[0-9]+}}, s4, s5
; CHECK-NEXT: fadd {{s[0-9]+}}, s6, s7
%1 = fadd float %a1, %a2
%2 = fadd float %a3, %a4
%3 = fadd float %a5, %a6
%4 = fadd float %a7, %a8
%5 = fadd float %1, %2
%6 = fadd float %3, %4
%7 = fadd float %5, %6
ret float %7
}
define double @call_arguments7(double %a1, double %a2, double %a3, double %a4, double %a5, double %a6, double %a7, double %a8) {
; CHECK-LABEL: call_arguments7
; CHECK: fadd {{d[0-9]+}}, d0, d1
; CHECK-NEXT: fadd {{d[0-9]+}}, d2, d3
; CHECK-NEXT: fadd {{d[0-9]+}}, d4, d5
; CHECK-NEXT: fadd {{d[0-9]+}}, d6, d7
%1 = fadd double %a1, %a2
%2 = fadd double %a3, %a4
%3 = fadd double %a5, %a6
%4 = fadd double %a7, %a8
%5 = fadd double %1, %2
%6 = fadd double %3, %4
%7 = fadd double %5, %6
ret double %7
}
define i64 @call_arguments8(i32 %a1, i64 %a2, i32 %a3, i64 %a4) {
; CHECK-LABEL: call_arguments8
; CHECK: ubfx [[REG1:x[0-9]+]], {{x[0-9]+}}, #0, #32
; CHECK: ubfx [[REG2:x[0-9]+]], {{x[0-9]+}}, #0, #32
; CHECK: add {{x[0-9]+}}, [[REG1]], x1
; CHECK-NEXT: add {{x[0-9]+}}, [[REG2]], x3
%aa1 = zext i32 %a1 to i64
%aa3 = zext i32 %a3 to i64
%1 = add i64 %aa1, %a2
%2 = add i64 %aa3, %a4
%3 = add i64 %1, %2
ret i64 %3
}
define void @call_arguments9(i8 %a1, i16 %a2, i32 %a3, i64 %a4, float %a5, double %a6, i64 %a7, double %a8) {
; CHECK-LABEL: call_arguments9
ret void
}
; Test that we use the correct register class for the branch.
define void @call_blr(i64 %Fn, i1 %c) {
; CHECK-LABEL: call_blr
; CHECK: blr
br i1 %c, label %bb1, label %bb2
bb1:
%1 = inttoptr i64 %Fn to void (i64)*
br label %bb2
bb2:
%2 = phi void (i64)* [ %1, %bb1 ], [ undef, %0 ]
call void %2(i64 1)
ret void
}