Skip to content

arm64: Replace RSH/RSZ -> CAST nodes with clearing register #121007

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
jonathandavies-arm wants to merge 9 commits into
base: main
Choose a base branch
from
Open

arm64: Replace RSH/RSZ -> CAST nodes with clearing register #121007

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
974f180
arm64: Replace RSH/RSZ -> CAST nodes with clearing register
jonathandavies-arm Oct 23, 2025
1e0e05e
Only optimise on unsigned casts
jonathandavies-arm Nov 12, 2025
6e345eb
Pass int into tests and perform casts in functions
jonathandavies-arm Nov 17, 2025
6aa1714
Merge branch 'main' into upstream/ce/right-shift-cast
jonathandavies-arm Nov 17, 2025
aed793e
Merge branch 'main' into upstream/ce/right-shift-cast
EgorBo Dec 3, 2025
2f5e093
Move optimisation from lowering to morph
jonathandavies-arm Jan 7, 2026
8f98e15
Merge branch 'main' into upstream/ce/right-shift-cast
jonathandavies-arm Jan 19, 2026
578d286
Add comment
jonathandavies-arm Jan 26, 2026
1927134
Merge branch 'main' into upstream/ce/right-shift-cast
jonathandavies-arm Jan 30, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
20 changes: 20 additions & 0 deletions src/coreclr/jit/morph.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11307,6 +11307,26 @@ GenTree* Compiler::fgMorphSmpOpOptional(GenTreeOp* tree, bool* optAssertionPropD

break;

case GT_RSH:
case GT_RSZ:
// Optimise right shift of an int -> unsigned cast by a constant amount, and the shift amount is >= the
// bit width of the cast to type. This always equals zero so replace shift & cast with setting to 0.
if (opts.OptimizationEnabled() && op1->OperIs(GT_CAST) && op2->IsCnsIntOrI())
{
GenTreeCast* cast = op1->AsCast();
GenTreeIntCon* cns = op2->AsIntCon();
if (!cast->gtOverflow() && cast->CastOp()->TypeIs(TYP_INT) && varTypeIsUnsigned(cast->CastToType()))
{
ssize_t shiftAmount = cns->IconValue();
Copy link
Contributor

@dhartglassMSFT dhartglassMSFT Jan 26, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can you add a small comment around here saying what the transform does? Just similar to the one on line 11279.

Change lgtm otherwise

Copy link
Contributor Author

@jonathandavies-arm jonathandavies-arm Jan 26, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Added a comment.

unsigned srcBits = genTypeSize(cast->CastToType()) * BITS_PER_BYTE;
if ((shiftAmount >= 0) && (static_cast<unsigned>(shiftAmount) >= srcBits))
{
return gtNewZeroConNode(tree->TypeGet());
}
}
}
break;

case GT_INIT_VAL:
// Initialization values for initBlk have special semantics - their lower
// byte is used to fill the struct. However, we allow 0 as a "bare" value,
Expand Down
349 changes: 349 additions & 0 deletions src/tests/JIT/opt/InstructionCombining/CastRightShift.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,349 @@
// Licensed to the .NET Foundation under one or more agreements.
// The .NET Foundation licenses this file to you under the MIT license.

using System;
using System.Runtime.CompilerServices;
using Xunit;

namespace TestCastRightShift
{
public class Program
{
[MethodImpl(MethodImplOptions.NoInlining)]
[Fact]
public static int CheckCastRightShift()
{
bool fail = false;

if (CastASR7_byte_int(255) == 0)
{
fail = true;
}

if (CastASR8_byte_int(255) != 0)
{
fail = true;
}

if (CastLSR7_byte_uint(255) != 1)
{
fail = true;
}

if (CastLSR8_byte_uint(255) != 0)
{
fail = true;
}

if (CastASR7_byte_long(255) == 0)
{
fail = true;
}

if (CastASR8_byte_long(255) != 0)
{
fail = true;
}

if (CastLSR7_byte_ulong(255) == 0)
{
fail = true;
}

if (CastLSR8_byte_ulong(255) != 0)
{
fail = true;
}

if (CastASR7_sbyte_int(-127) != -1)
{
fail = true;
}

if (CastASR8_sbyte_int(-127) != -1)
{
fail = true;
}

if (CastASR7_sbyte_long(-127) != -1)
{
fail = true;
}

if (CastASR8_sbyte_long(-127) != -1)
{
fail = true;
}

if (CastASR15_ushort_int(65535) == 0)
{
fail = true;
}

if (CastASR16_ushort_int(65535) != 0)
{
fail = true;
}

if (CastLSR15_ushort_uint(65535) == 0)
{
fail = true;
}

if (CastLSR16_ushort_uint(65535) != 0)
{
fail = true;
}

if (CastASR15_ushort_long(65535) == 0)
{
fail = true;
}

if (CastASR16_ushort_long(65535) != 0)
{
fail = true;
}

if (CastLSR15_ushort_ulong(65535) == 0)
{
fail = true;
}

if (CastLSR16_ushort_ulong(65535) != 0)
{
fail = true;
}

if (CastASR15_short_int(-1) != -1)
{
fail = true;
}

if (CastASR16_short_int(-1) != -1)
{
fail = true;
}

if (CastASR15_short_long(-1) != -1)
{
fail = true;
}

if (CastASR16_short_long(-1) != -1)
{
fail = true;
}

if (fail)
{
return 101;
}
return 100;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static int CastASR7_byte_int(int x)
{
//ARM64-FULL-LINE: uxtb {{w[0-9]+}}, {{w[0-9]+}}
//ARM64-FULL-LINE: asr {{w[0-9]+}}, {{w[0-9]+}}, #7
return (byte)x >> 7;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static int CastASR8_byte_int(int x)
{
//ARM64-FULL-LINE: mov {{w[0-9]+}}, wzr
return (byte)x >> 8;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static uint CastLSR7_byte_uint(int x)
{
//ARM64-FULL-LINE: uxtb {{w[0-9]+}}, {{w[0-9]+}}
//ARM64-FULL-LINE: lsr {{w[0-9]+}}, {{w[0-9]+}}, #7
return (uint)(byte)x >> 7;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static uint CastLSR8_byte_uint(int x)
{
//ARM64-FULL-LINE: mov {{w[0-9]+}}, wzr
return (uint)(byte)x >> 8;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static long CastASR7_byte_long(int x)
{
//ARM64-FULL-LINE: uxtb {{w[0-9]+}}, {{w[0-9]+}}
//ARM64-FULL-LINE: asr {{w[0-9]+}}, {{w[0-9]+}}, #7
//ARM64-FULL-LINE: sxtw {{x[0-9]+}}, {{w[0-9]+}}
return (byte)x >> 7;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static long CastASR8_byte_long(int x)
{
//ARM64-FULL-LINE: mov {{x[0-9]+}}, xzr
return (byte)x >> 8;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static ulong CastLSR7_byte_ulong(int x)
{
//ARM64-FULL-LINE: uxtb {{w[0-9]+}}, {{w[0-9]+}}
//ARM64-FULL-LINE: mov {{w[0-9]+}}, {{w[0-9]+}}
//ARM64-FULL-LINE: lsr {{x[0-9]+}}, {{x[0-9]+}}, #7
return (ulong)(byte)x >> 7;
}

// This produces the following tree section with 2 CASTs
// We should check all the CASTs and then check that the lsr value
// is greater than or equal to smallest src bits. In this case
// we could change to mov w0, wzr.
// * RSZ long
// +--* CAST long <- ulong
// | \--* CAST int <- ubyte <- int
// | \--* LCL_VAR int
[MethodImpl(MethodImplOptions.NoInlining)]
static ulong CastLSR8_byte_ulong(int x)
{
//ARM64-FULL-LINE: uxtb {{w[0-9]+}}, {{w[0-9]+}}
//ARM64-FULL-LINE: mov {{w[0-9]+}}, {{w[0-9]+}}
//ARM64-FULL-LINE: lsr {{x[0-9]+}}, {{x[0-9]+}}, #8
return (ulong)(byte)x >> 8;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static int CastASR7_sbyte_int(int x)
{
//ARM64-FULL-LINE: sxtb {{w[0-9]+}}, {{w[0-9]+}}
//ARM64-FULL-LINE: asr {{w[0-9]+}}, {{w[0-9]+}}, #7
return (sbyte)x >> 7;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static int CastASR8_sbyte_int(int x)
{
//ARM64-FULL-LINE: sxtb {{w[0-9]+}}, {{w[0-9]+}}
//ARM64-FULL-LINE: asr {{w[0-9]+}}, {{w[0-9]+}}, #8
return (sbyte)x >> 8;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static long CastASR7_sbyte_long(int x)
{
//ARM64-FULL-LINE: sxtb {{w[0-9]+}}, {{w[0-9]+}}
//ARM64-FULL-LINE: asr {{w[0-9]+}}, {{w[0-9]+}}, #7
//ARM64-FULL-LINE: sxtw {{x[0-9]+}}, {{w[0-9]+}}
return (sbyte)x >> 7;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static long CastASR8_sbyte_long(int x)
{
//ARM64-FULL-LINE: sxtb {{w[0-9]+}}, {{w[0-9]+}}
//ARM64-FULL-LINE: asr {{w[0-9]+}}, {{w[0-9]+}}, #8
//ARM64-FULL-LINE: sxtw {{x[0-9]+}}, {{w[0-9]+}}
return (sbyte)x >> 8;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static int CastASR15_ushort_int(int x)
{
//ARM64-FULL-LINE: uxth {{w[0-9]+}}, {{w[0-9]+}}
//ARM64-FULL-LINE: asr {{w[0-9]+}}, {{w[0-9]+}}, #15
return (ushort)x >> 15;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static int CastASR16_ushort_int(int x)
{
//ARM64-FULL-LINE: mov {{w[0-9]+}}, wzr
return (ushort)x >> 16;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static uint CastLSR15_ushort_uint(int x)
{
//ARM64-FULL-LINE: uxth {{w[0-9]+}}, {{w[0-9]+}}
//ARM64-FULL-LINE: lsr {{w[0-9]+}}, {{w[0-9]+}}, #15
return (uint)(ushort)x >> 15;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static uint CastLSR16_ushort_uint(int x)
{
//ARM64-FULL-LINE: mov {{w[0-9]+}}, wzr
return (uint)(ushort)x >> 16;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static long CastASR15_ushort_long(int x)
{
//ARM64-FULL-LINE: uxth {{w[0-9]+}}, {{w[0-9]+}}
//ARM64-FULL-LINE: asr {{w[0-9]+}}, {{w[0-9]+}}, #15
//ARM64-FULL-LINE: sxtw {{x[0-9]+}}, {{w[0-9]+}}
return (ushort)x >> 15;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static long CastASR16_ushort_long(int x)
{
//ARM64-FULL-LINE: mov {{x[0-9]+}}, xzr
return (ushort)x >> 16;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static ulong CastLSR15_ushort_ulong(int x)
{
//ARM64-FULL-LINE: uxth {{w[0-9]+}}, {{w[0-9]+}}
//ARM64-FULL-LINE: mov {{w[0-9]+}}, {{w[0-9]+}}
//ARM64-FULL-LINE: lsr {{x[0-9]+}}, {{x[0-9]+}}, #15
return (ulong)(ushort)x >> 15;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static ulong CastLSR16_ushort_ulong(int x)
{
//ARM64-FULL-LINE: uxth {{w[0-9]+}}, {{w[0-9]+}}
//ARM64-FULL-LINE: mov {{w[0-9]+}}, {{w[0-9]+}}
//ARM64-FULL-LINE: lsr {{x[0-9]+}}, {{x[0-9]+}}, #16
return (ulong)(ushort)x >> 16;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static int CastASR15_short_int(int x)
{
//ARM64-FULL-LINE: sxth {{w[0-9]+}}, {{w[0-9]+}}
//ARM64-FULL-LINE: asr {{w[0-9]+}}, {{w[0-9]+}}, #15
return (short)x >> 15;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static int CastASR16_short_int(int x)
{
//ARM64-FULL-LINE: sxth {{w[0-9]+}}, {{w[0-9]+}}
//ARM64-FULL-LINE: asr {{w[0-9]+}}, {{w[0-9]+}}, #16
return (short)x >> 16;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static long CastASR15_short_long(int x)
{
//ARM64-FULL-LINE: sxth {{w[0-9]+}}, {{w[0-9]+}}
//ARM64-FULL-LINE: asr {{w[0-9]+}}, {{w[0-9]+}}, #15
//ARM64-FULL-LINE: sxtw {{x[0-9]+}}, {{w[0-9]+}}
return (short)x >> 15;
}

[MethodImpl(MethodImplOptions.NoInlining)]
static long CastASR16_short_long(int x)
{
//ARM64-FULL-LINE: sxth {{w[0-9]+}}, {{w[0-9]+}}
//ARM64-FULL-LINE: asr {{w[0-9]+}}, {{w[0-9]+}}, #16
//ARM64-FULL-LINE: sxtw {{x[0-9]+}}, {{w[0-9]+}}
return (short)x >> 16;
}
}
}
17 changes: 17 additions & 0 deletions src/tests/JIT/opt/InstructionCombining/CastRightShift.csproj
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,17 @@
<Project Sdk="Microsoft.NET.Sdk">
<PropertyGroup>
<!-- Needed for CLRTestEnvironmentVariable -->
<RequiresProcessIsolation>true</RequiresProcessIsolation>
</PropertyGroup>
<PropertyGroup>
<DebugType>None</DebugType>
<Optimize>True</Optimize>
</PropertyGroup>
<ItemGroup>
<Compile Include="CastRightShift.cs">
<HasDisasmCheck>true</HasDisasmCheck>
</Compile>
<CLRTestEnvironmentVariable Include="DOTNET_TieredCompilation" Value="0" />
<CLRTestEnvironmentVariable Include="DOTNET_JITMinOpts" Value="0" />
</ItemGroup>
</Project>
Loading