[llvm] [AMDGPU] Drop high 32 bits of aperture registers (PR #158725)

[Stanislav Mekhanoshin via llvm-commits]

================
@@ -1225,7 +1233,7 @@ def SSrc_bf16: SrcRegOrImm9 <SReg_32, "OPERAND_REG_IMM_BF16">;
 def SSrc_f16 : SrcRegOrImm9 <SReg_32, "OPERAND_REG_IMM_FP16">;
 def SSrc_b32 : SrcRegOrImm9 <SReg_32, "OPERAND_REG_IMM_INT32">;
 def SSrc_f32 : SrcRegOrImm9 <SReg_32, "OPERAND_REG_IMM_FP32">;
-def SSrc_b64 : SrcRegOrImm9 <SReg_64, "OPERAND_REG_IMM_INT64">;
+def SSrc_b64 : SrcRegOrImm9 <SReg_64_Plus, "OPERAND_REG_IMM_INT64">;
----------------
rampitec wrote:

These tests are incredibly stupid, and I think it does not do anything, but if it is needed to unblock this, then: https://github.com/llvm/llvm-project/pull/158823. I still insist it shall not be a part of this PR.

https://github.com/llvm/llvm-project/pull/158725