llvm-project/llvm/test/CodeGen/SystemZ/fp-strict-conv-10.ll

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

92 lines
3.0 KiB
LLVM
Raw Normal View History

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; Test strict conversion of floating-point values to unsigned i32s (z10 only).
;
; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z10 | FileCheck %s
; z10 doesn't have native support for unsigned fp-to-i32 conversions;
; they were added in z196 as the Convert to Logical family of instructions.
; Promoting to i64 doesn't generate an inexact condition for values that are
; outside the i32 range but in the i64 range, so use the default expansion.
; Note that the strict expansion sequence must be used.
declare i32 @llvm.experimental.constrained.fptoui.i32.f32(float, metadata)
declare i32 @llvm.experimental.constrained.fptoui.i32.f64(double, metadata)
declare i32 @llvm.experimental.constrained.fptoui.i32.f128(fp128, metadata)
; Test f32->i32.
define i32 @f1(float %f) #0 {
; CHECK-LABEL: f1:
; CHECK: # %bb.0:
; CHECK-NEXT: larl %r1, .LCPI0_0
; CHECK-NEXT: le %f1, 0(%r1)
; CHECK-NEXT: kebr %f0, %f1
[TargetLowering] Fix another potential FPE in expandFP_TO_UINT D53794 introduced code to perform the FP_TO_UINT expansion via FP_TO_SINT in a way that would never expose floating-point exceptions in the intermediate steps. Unfortunately, I just noticed there is still a way this can happen. As discussed in D53794, the compiler now generates this sequence: // Sel = Src < 0x8000000000000000 // Val = select Sel, Src, Src - 0x8000000000000000 // Ofs = select Sel, 0, 0x8000000000000000 // Result = fp_to_sint(Val) ^ Ofs The problem is with the Src - 0x8000000000000000 expression. As I mentioned in the original review, that expression can never overflow or underflow if the original value is in range for FP_TO_UINT. But I missed that we can get an Inexact exception in the case where Src is a very small positive value. (In this case the result of the sub is ignored, but that doesn't help.) Instead, I'd suggest to use the following sequence: // Sel = Src < 0x8000000000000000 // FltOfs = select Sel, 0, 0x8000000000000000 // IntOfs = select Sel, 0, 0x8000000000000000 // Result = fp_to_sint(Val - FltOfs) ^ IntOfs In the case where the value is already in range of FP_TO_SINT, we now simply compute Val - 0, which now definitely cannot trap (unless Val is a NaN in which case we'd want to trap anyway). In the case where the value is not in range of FP_TO_SINT, but still in range of FP_TO_UINT, the sub can never be inexact, as Val is between 2^(n-1) and (2^n)-1, i.e. always has the 2^(n-1) bit set, and the sub is always simply clearing that bit. There is a slight complication in the case where Val is a constant, so we know at compile time whether Sel is true or false. In that scenario, the old code would automatically optimize the sub away, while this no longer happens with the new code. Instead, I've added extra code to check for this case and then just fall back to FP_TO_SINT directly. (This seems to catch even slightly more cases.) Original version of the patch by Ulrich Weigand. X86 changes added by Craig Topper Differential Revision: https://reviews.llvm.org/D67105
2019-12-07 06:11:04 +08:00
; CHECK-NEXT: jnl .LBB0_2
; CHECK-NEXT: # %bb.1:
[TargetLowering] Fix another potential FPE in expandFP_TO_UINT D53794 introduced code to perform the FP_TO_UINT expansion via FP_TO_SINT in a way that would never expose floating-point exceptions in the intermediate steps. Unfortunately, I just noticed there is still a way this can happen. As discussed in D53794, the compiler now generates this sequence: // Sel = Src < 0x8000000000000000 // Val = select Sel, Src, Src - 0x8000000000000000 // Ofs = select Sel, 0, 0x8000000000000000 // Result = fp_to_sint(Val) ^ Ofs The problem is with the Src - 0x8000000000000000 expression. As I mentioned in the original review, that expression can never overflow or underflow if the original value is in range for FP_TO_UINT. But I missed that we can get an Inexact exception in the case where Src is a very small positive value. (In this case the result of the sub is ignored, but that doesn't help.) Instead, I'd suggest to use the following sequence: // Sel = Src < 0x8000000000000000 // FltOfs = select Sel, 0, 0x8000000000000000 // IntOfs = select Sel, 0, 0x8000000000000000 // Result = fp_to_sint(Val - FltOfs) ^ IntOfs In the case where the value is already in range of FP_TO_SINT, we now simply compute Val - 0, which now definitely cannot trap (unless Val is a NaN in which case we'd want to trap anyway). In the case where the value is not in range of FP_TO_SINT, but still in range of FP_TO_UINT, the sub can never be inexact, as Val is between 2^(n-1) and (2^n)-1, i.e. always has the 2^(n-1) bit set, and the sub is always simply clearing that bit. There is a slight complication in the case where Val is a constant, so we know at compile time whether Sel is true or false. In that scenario, the old code would automatically optimize the sub away, while this no longer happens with the new code. Instead, I've added extra code to check for this case and then just fall back to FP_TO_SINT directly. (This seems to catch even slightly more cases.) Original version of the patch by Ulrich Weigand. X86 changes added by Craig Topper Differential Revision: https://reviews.llvm.org/D67105
2019-12-07 06:11:04 +08:00
; CHECK-NEXT: lhi %r0, 0
; CHECK-NEXT: lzer %f1
; CHECK-NEXT: j .LBB0_3
; CHECK-NEXT: .LBB0_2:
[TargetLowering] Fix another potential FPE in expandFP_TO_UINT D53794 introduced code to perform the FP_TO_UINT expansion via FP_TO_SINT in a way that would never expose floating-point exceptions in the intermediate steps. Unfortunately, I just noticed there is still a way this can happen. As discussed in D53794, the compiler now generates this sequence: // Sel = Src < 0x8000000000000000 // Val = select Sel, Src, Src - 0x8000000000000000 // Ofs = select Sel, 0, 0x8000000000000000 // Result = fp_to_sint(Val) ^ Ofs The problem is with the Src - 0x8000000000000000 expression. As I mentioned in the original review, that expression can never overflow or underflow if the original value is in range for FP_TO_UINT. But I missed that we can get an Inexact exception in the case where Src is a very small positive value. (In this case the result of the sub is ignored, but that doesn't help.) Instead, I'd suggest to use the following sequence: // Sel = Src < 0x8000000000000000 // FltOfs = select Sel, 0, 0x8000000000000000 // IntOfs = select Sel, 0, 0x8000000000000000 // Result = fp_to_sint(Val - FltOfs) ^ IntOfs In the case where the value is already in range of FP_TO_SINT, we now simply compute Val - 0, which now definitely cannot trap (unless Val is a NaN in which case we'd want to trap anyway). In the case where the value is not in range of FP_TO_SINT, but still in range of FP_TO_UINT, the sub can never be inexact, as Val is between 2^(n-1) and (2^n)-1, i.e. always has the 2^(n-1) bit set, and the sub is always simply clearing that bit. There is a slight complication in the case where Val is a constant, so we know at compile time whether Sel is true or false. In that scenario, the old code would automatically optimize the sub away, while this no longer happens with the new code. Instead, I've added extra code to check for this case and then just fall back to FP_TO_SINT directly. (This seems to catch even slightly more cases.) Original version of the patch by Ulrich Weigand. X86 changes added by Craig Topper Differential Revision: https://reviews.llvm.org/D67105
2019-12-07 06:11:04 +08:00
; CHECK-NEXT: llilh %r0, 32768
; CHECK-NEXT: .LBB0_3:
; CHECK-NEXT: sebr %f0, %f1
; CHECK-NEXT: cfebr %r2, 5, %f0
; CHECK-NEXT: xr %r2, %r0
; CHECK-NEXT: br %r14
%conv = call i32 @llvm.experimental.constrained.fptoui.i32.f32(float %f,
metadata !"fpexcept.strict") #0
ret i32 %conv
}
; Test f64->i32.
define i32 @f2(double %f) #0 {
; CHECK-LABEL: f2:
; CHECK: # %bb.0:
; CHECK-NEXT: larl %r1, .LCPI1_0
; CHECK-NEXT: ldeb %f1, 0(%r1)
; CHECK-NEXT: kdbr %f0, %f1
[TargetLowering] Fix another potential FPE in expandFP_TO_UINT D53794 introduced code to perform the FP_TO_UINT expansion via FP_TO_SINT in a way that would never expose floating-point exceptions in the intermediate steps. Unfortunately, I just noticed there is still a way this can happen. As discussed in D53794, the compiler now generates this sequence: // Sel = Src < 0x8000000000000000 // Val = select Sel, Src, Src - 0x8000000000000000 // Ofs = select Sel, 0, 0x8000000000000000 // Result = fp_to_sint(Val) ^ Ofs The problem is with the Src - 0x8000000000000000 expression. As I mentioned in the original review, that expression can never overflow or underflow if the original value is in range for FP_TO_UINT. But I missed that we can get an Inexact exception in the case where Src is a very small positive value. (In this case the result of the sub is ignored, but that doesn't help.) Instead, I'd suggest to use the following sequence: // Sel = Src < 0x8000000000000000 // FltOfs = select Sel, 0, 0x8000000000000000 // IntOfs = select Sel, 0, 0x8000000000000000 // Result = fp_to_sint(Val - FltOfs) ^ IntOfs In the case where the value is already in range of FP_TO_SINT, we now simply compute Val - 0, which now definitely cannot trap (unless Val is a NaN in which case we'd want to trap anyway). In the case where the value is not in range of FP_TO_SINT, but still in range of FP_TO_UINT, the sub can never be inexact, as Val is between 2^(n-1) and (2^n)-1, i.e. always has the 2^(n-1) bit set, and the sub is always simply clearing that bit. There is a slight complication in the case where Val is a constant, so we know at compile time whether Sel is true or false. In that scenario, the old code would automatically optimize the sub away, while this no longer happens with the new code. Instead, I've added extra code to check for this case and then just fall back to FP_TO_SINT directly. (This seems to catch even slightly more cases.) Original version of the patch by Ulrich Weigand. X86 changes added by Craig Topper Differential Revision: https://reviews.llvm.org/D67105
2019-12-07 06:11:04 +08:00
; CHECK-NEXT: jnl .LBB1_2
; CHECK-NEXT: # %bb.1:
[TargetLowering] Fix another potential FPE in expandFP_TO_UINT D53794 introduced code to perform the FP_TO_UINT expansion via FP_TO_SINT in a way that would never expose floating-point exceptions in the intermediate steps. Unfortunately, I just noticed there is still a way this can happen. As discussed in D53794, the compiler now generates this sequence: // Sel = Src < 0x8000000000000000 // Val = select Sel, Src, Src - 0x8000000000000000 // Ofs = select Sel, 0, 0x8000000000000000 // Result = fp_to_sint(Val) ^ Ofs The problem is with the Src - 0x8000000000000000 expression. As I mentioned in the original review, that expression can never overflow or underflow if the original value is in range for FP_TO_UINT. But I missed that we can get an Inexact exception in the case where Src is a very small positive value. (In this case the result of the sub is ignored, but that doesn't help.) Instead, I'd suggest to use the following sequence: // Sel = Src < 0x8000000000000000 // FltOfs = select Sel, 0, 0x8000000000000000 // IntOfs = select Sel, 0, 0x8000000000000000 // Result = fp_to_sint(Val - FltOfs) ^ IntOfs In the case where the value is already in range of FP_TO_SINT, we now simply compute Val - 0, which now definitely cannot trap (unless Val is a NaN in which case we'd want to trap anyway). In the case where the value is not in range of FP_TO_SINT, but still in range of FP_TO_UINT, the sub can never be inexact, as Val is between 2^(n-1) and (2^n)-1, i.e. always has the 2^(n-1) bit set, and the sub is always simply clearing that bit. There is a slight complication in the case where Val is a constant, so we know at compile time whether Sel is true or false. In that scenario, the old code would automatically optimize the sub away, while this no longer happens with the new code. Instead, I've added extra code to check for this case and then just fall back to FP_TO_SINT directly. (This seems to catch even slightly more cases.) Original version of the patch by Ulrich Weigand. X86 changes added by Craig Topper Differential Revision: https://reviews.llvm.org/D67105
2019-12-07 06:11:04 +08:00
; CHECK-NEXT: lhi %r0, 0
; CHECK-NEXT: lzdr %f1
; CHECK-NEXT: j .LBB1_3
; CHECK-NEXT: .LBB1_2:
[TargetLowering] Fix another potential FPE in expandFP_TO_UINT D53794 introduced code to perform the FP_TO_UINT expansion via FP_TO_SINT in a way that would never expose floating-point exceptions in the intermediate steps. Unfortunately, I just noticed there is still a way this can happen. As discussed in D53794, the compiler now generates this sequence: // Sel = Src < 0x8000000000000000 // Val = select Sel, Src, Src - 0x8000000000000000 // Ofs = select Sel, 0, 0x8000000000000000 // Result = fp_to_sint(Val) ^ Ofs The problem is with the Src - 0x8000000000000000 expression. As I mentioned in the original review, that expression can never overflow or underflow if the original value is in range for FP_TO_UINT. But I missed that we can get an Inexact exception in the case where Src is a very small positive value. (In this case the result of the sub is ignored, but that doesn't help.) Instead, I'd suggest to use the following sequence: // Sel = Src < 0x8000000000000000 // FltOfs = select Sel, 0, 0x8000000000000000 // IntOfs = select Sel, 0, 0x8000000000000000 // Result = fp_to_sint(Val - FltOfs) ^ IntOfs In the case where the value is already in range of FP_TO_SINT, we now simply compute Val - 0, which now definitely cannot trap (unless Val is a NaN in which case we'd want to trap anyway). In the case where the value is not in range of FP_TO_SINT, but still in range of FP_TO_UINT, the sub can never be inexact, as Val is between 2^(n-1) and (2^n)-1, i.e. always has the 2^(n-1) bit set, and the sub is always simply clearing that bit. There is a slight complication in the case where Val is a constant, so we know at compile time whether Sel is true or false. In that scenario, the old code would automatically optimize the sub away, while this no longer happens with the new code. Instead, I've added extra code to check for this case and then just fall back to FP_TO_SINT directly. (This seems to catch even slightly more cases.) Original version of the patch by Ulrich Weigand. X86 changes added by Craig Topper Differential Revision: https://reviews.llvm.org/D67105
2019-12-07 06:11:04 +08:00
; CHECK-NEXT: llilh %r0, 32768
; CHECK-NEXT: .LBB1_3:
; CHECK-NEXT: sdbr %f0, %f1
; CHECK-NEXT: cfdbr %r2, 5, %f0
; CHECK-NEXT: xr %r2, %r0
; CHECK-NEXT: br %r14
%conv = call i32 @llvm.experimental.constrained.fptoui.i32.f64(double %f,
metadata !"fpexcept.strict") #0
ret i32 %conv
}
; Test f128->i32.
define i32 @f3(fp128 *%src) #0 {
; CHECK-LABEL: f3:
; CHECK: # %bb.0:
; CHECK-NEXT: ld %f0, 0(%r2)
; CHECK-NEXT: ld %f2, 8(%r2)
; CHECK-NEXT: larl %r1, .LCPI2_0
; CHECK-NEXT: lxeb %f1, 0(%r1)
; CHECK-NEXT: kxbr %f0, %f1
[TargetLowering] Fix another potential FPE in expandFP_TO_UINT D53794 introduced code to perform the FP_TO_UINT expansion via FP_TO_SINT in a way that would never expose floating-point exceptions in the intermediate steps. Unfortunately, I just noticed there is still a way this can happen. As discussed in D53794, the compiler now generates this sequence: // Sel = Src < 0x8000000000000000 // Val = select Sel, Src, Src - 0x8000000000000000 // Ofs = select Sel, 0, 0x8000000000000000 // Result = fp_to_sint(Val) ^ Ofs The problem is with the Src - 0x8000000000000000 expression. As I mentioned in the original review, that expression can never overflow or underflow if the original value is in range for FP_TO_UINT. But I missed that we can get an Inexact exception in the case where Src is a very small positive value. (In this case the result of the sub is ignored, but that doesn't help.) Instead, I'd suggest to use the following sequence: // Sel = Src < 0x8000000000000000 // FltOfs = select Sel, 0, 0x8000000000000000 // IntOfs = select Sel, 0, 0x8000000000000000 // Result = fp_to_sint(Val - FltOfs) ^ IntOfs In the case where the value is already in range of FP_TO_SINT, we now simply compute Val - 0, which now definitely cannot trap (unless Val is a NaN in which case we'd want to trap anyway). In the case where the value is not in range of FP_TO_SINT, but still in range of FP_TO_UINT, the sub can never be inexact, as Val is between 2^(n-1) and (2^n)-1, i.e. always has the 2^(n-1) bit set, and the sub is always simply clearing that bit. There is a slight complication in the case where Val is a constant, so we know at compile time whether Sel is true or false. In that scenario, the old code would automatically optimize the sub away, while this no longer happens with the new code. Instead, I've added extra code to check for this case and then just fall back to FP_TO_SINT directly. (This seems to catch even slightly more cases.) Original version of the patch by Ulrich Weigand. X86 changes added by Craig Topper Differential Revision: https://reviews.llvm.org/D67105
2019-12-07 06:11:04 +08:00
; CHECK-NEXT: jnl .LBB2_2
; CHECK-NEXT: # %bb.1:
[TargetLowering] Fix another potential FPE in expandFP_TO_UINT D53794 introduced code to perform the FP_TO_UINT expansion via FP_TO_SINT in a way that would never expose floating-point exceptions in the intermediate steps. Unfortunately, I just noticed there is still a way this can happen. As discussed in D53794, the compiler now generates this sequence: // Sel = Src < 0x8000000000000000 // Val = select Sel, Src, Src - 0x8000000000000000 // Ofs = select Sel, 0, 0x8000000000000000 // Result = fp_to_sint(Val) ^ Ofs The problem is with the Src - 0x8000000000000000 expression. As I mentioned in the original review, that expression can never overflow or underflow if the original value is in range for FP_TO_UINT. But I missed that we can get an Inexact exception in the case where Src is a very small positive value. (In this case the result of the sub is ignored, but that doesn't help.) Instead, I'd suggest to use the following sequence: // Sel = Src < 0x8000000000000000 // FltOfs = select Sel, 0, 0x8000000000000000 // IntOfs = select Sel, 0, 0x8000000000000000 // Result = fp_to_sint(Val - FltOfs) ^ IntOfs In the case where the value is already in range of FP_TO_SINT, we now simply compute Val - 0, which now definitely cannot trap (unless Val is a NaN in which case we'd want to trap anyway). In the case where the value is not in range of FP_TO_SINT, but still in range of FP_TO_UINT, the sub can never be inexact, as Val is between 2^(n-1) and (2^n)-1, i.e. always has the 2^(n-1) bit set, and the sub is always simply clearing that bit. There is a slight complication in the case where Val is a constant, so we know at compile time whether Sel is true or false. In that scenario, the old code would automatically optimize the sub away, while this no longer happens with the new code. Instead, I've added extra code to check for this case and then just fall back to FP_TO_SINT directly. (This seems to catch even slightly more cases.) Original version of the patch by Ulrich Weigand. X86 changes added by Craig Topper Differential Revision: https://reviews.llvm.org/D67105
2019-12-07 06:11:04 +08:00
; CHECK-NEXT: lhi %r0, 0
; CHECK-NEXT: lzxr %f1
; CHECK-NEXT: j .LBB2_3
; CHECK-NEXT: .LBB2_2:
[TargetLowering] Fix another potential FPE in expandFP_TO_UINT D53794 introduced code to perform the FP_TO_UINT expansion via FP_TO_SINT in a way that would never expose floating-point exceptions in the intermediate steps. Unfortunately, I just noticed there is still a way this can happen. As discussed in D53794, the compiler now generates this sequence: // Sel = Src < 0x8000000000000000 // Val = select Sel, Src, Src - 0x8000000000000000 // Ofs = select Sel, 0, 0x8000000000000000 // Result = fp_to_sint(Val) ^ Ofs The problem is with the Src - 0x8000000000000000 expression. As I mentioned in the original review, that expression can never overflow or underflow if the original value is in range for FP_TO_UINT. But I missed that we can get an Inexact exception in the case where Src is a very small positive value. (In this case the result of the sub is ignored, but that doesn't help.) Instead, I'd suggest to use the following sequence: // Sel = Src < 0x8000000000000000 // FltOfs = select Sel, 0, 0x8000000000000000 // IntOfs = select Sel, 0, 0x8000000000000000 // Result = fp_to_sint(Val - FltOfs) ^ IntOfs In the case where the value is already in range of FP_TO_SINT, we now simply compute Val - 0, which now definitely cannot trap (unless Val is a NaN in which case we'd want to trap anyway). In the case where the value is not in range of FP_TO_SINT, but still in range of FP_TO_UINT, the sub can never be inexact, as Val is between 2^(n-1) and (2^n)-1, i.e. always has the 2^(n-1) bit set, and the sub is always simply clearing that bit. There is a slight complication in the case where Val is a constant, so we know at compile time whether Sel is true or false. In that scenario, the old code would automatically optimize the sub away, while this no longer happens with the new code. Instead, I've added extra code to check for this case and then just fall back to FP_TO_SINT directly. (This seems to catch even slightly more cases.) Original version of the patch by Ulrich Weigand. X86 changes added by Craig Topper Differential Revision: https://reviews.llvm.org/D67105
2019-12-07 06:11:04 +08:00
; CHECK-NEXT: llilh %r0, 32768
; CHECK-NEXT: .LBB2_3:
; CHECK-NEXT: sxbr %f0, %f1
; CHECK-NEXT: cfxbr %r2, 5, %f0
; CHECK-NEXT: xr %r2, %r0
; CHECK-NEXT: br %r14
%f = load fp128, fp128 *%src
%conv = call i32 @llvm.experimental.constrained.fptoui.i32.f128(fp128 %f,
metadata !"fpexcept.strict") #0
ret i32 %conv
}
attributes #0 = { strictfp }