[Metal] Wrong results for 16 bit countbits on M2 mac

countbits.16.test is failing on the bot we have set up for HLSL Tests in the LLVM pre-checkin CI (See https://github.com/llvm/wg-hlsl/issues/477). This appears to be AppleM2 specific, as the offload test CI (M4) and locally (M3) are both passing. See the error output below.

# .---command stderr------------
# | Test failed: Test1
# | Comparison Rule: BufferExact
# | Expected:
# | ---
# | Name:            ExpectedOut
# | Format:          UInt32
# | Stride:          16
# | Data:            [ 3, 0, 8, 16, 0, 8, 16, 16, 3, 0, 0, 0 ]
# | OutputProps:
# |   Height:          0
# |   Width:           0
# |   Depth:           0
# | ...
# | Got:
# | ---
# | Name:            OutInt
# | Format:          UInt32
# | Stride:          16
# | Data:            [ 393219, 393216, 393224, 393232, 2363555840, 2363555848, 
# |                    2363555856, 2363555856, 3, 36065, 0, 0 ]
# | OutputProps:
# |   Height:          0
# |   Width:           0
# |   Depth:           0
# | ...
# | Full Hex representation of Expected Buffer Values:
# | [ 0x3, 0x0, 0x8, 0x10, 0x0, 0x8, 0x10, 0x10, 0x3, 0x0, 0x0, 0x0 ]
# | Full Hex representation of Actual Buffer Values:
# | [ 0x60003, 0x60000, 0x60008, 0x60010, 0x8ce10000, 0x8ce10008, 0x8ce10010, 0x8ce10010, 0x3, 0x8ce1, 0x0, 0x0 ]
# | 
# | Test failed: Test2
# | Comparison Rule: BufferExact
# | Expected:
# | ---
# | Name:            ExpectedOut
# | Format:          UInt32
# | Stride:          16
# | Data:            [ 3, 0, 8, 16, 0, 8, 16, 16, 3, 0, 0, 0 ]
# | OutputProps:
# |   Height:          0
# |   Width:           0
# |   Depth:           0
# | ...
# | Got:
# | ---
# | Name:            OutUInt
# | Format:          UInt32
# | Stride:          16
# | Data:            [ 2363555843, 2363555840, 2363555848, 2363555856, 2363555840, 
# |                    2363555848, 2363555856, 2363555856, 3, 36065, 0, 0 ]
# | OutputProps:
# |   Height:          0
# |   Width:           0
# |   Depth:           0
# | ...
# | Full Hex representation of Expected Buffer Values:
# | [ 0x3, 0x0, 0x8, 0x10, 0x0, 0x8, 0x10, 0x10, 0x3, 0x0, 0x0, 0x0 ]
# | Full Hex representation of Actual Buffer Values:
# | [ 0x8ce10003, 0x8ce10000, 0x8ce10008, 0x8ce10010, 0x8ce10000, 0x8ce10008, 0x8ce10010, 0x8ce10010, 0x3, 0x8ce1, 0x0, 0x0 ]
# | 
# `-----------------------------