AMDGPU: Fix assembler asserting on expressions in vop3 instructions

[arsenm]

Fixes #100075

[arsenm]

• AMDGPU: Fix assembler asserting on expressions in vop3 instructions #100103 👈
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

Join @arsenm and the rest of your teammates on Graphite

[llvmbot]

@llvm/pr-subscribers-mc

@llvm/pr-subscribers-backend-amdgpu

Author: Matt Arsenault (arsenm)

Changes

Fixes #100075

---

Full diff: https://github.com/llvm/llvm-project/pull/100103.diff

2 Files Affected:

• (modified) llvm/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp (+1-1)
• (added) llvm/test/MC/AMDGPU/reloc-operands-gfx10.s (+31)

diff --git a/llvm/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp b/llvm/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp
index 92c3b26ca4d6f..9616a0ff57134 100644
--- a/llvm/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp
+++ b/llvm/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp
@@ -8649,7 +8649,7 @@ void AMDGPUAsmParser::cvtVOP3(MCInst &Inst, const OperandVector &Operands,
       Op.addRegOrImmWithFPInputModsOperands(Inst, 2);
     } else if (Op.isImmModifier()) {
       OptionalIdx[Op.getImmTy()] = I;
-    } else if (Op.isRegOrImm()) {
+    } else if (Op.isRegOrImm() || Op.isExpr()) {
       Op.addRegOrImmOperands(Inst, 1);
     } else {
       llvm_unreachable("unhandled operand type");
diff --git a/llvm/test/MC/AMDGPU/reloc-operands-gfx10.s b/llvm/test/MC/AMDGPU/reloc-operands-gfx10.s
new file mode 100644
index 0000000000000..b0fe71ddadb3b
--- /dev/null
+++ b/llvm/test/MC/AMDGPU/reloc-operands-gfx10.s
@@ -0,0 +1,31 @@
+// RUN: llvm-mc -triple=amdgcn -mcpu=gfx1010 -show-encoding %s | FileCheck -check-prefix=GFX10 %s
+
+; test vop3 operands
+
+// GFX10: v_mad_u32_u24 v0, g0@abs32@lo, v0, 12   ; encoding: [0x00,0x00,0x43,0xd5,0xff,0x00,0x32,0x02,A,A,A,A]
+// GFX10-NEXT:                                    ;   fixup A - offset: 8, value: g0@abs32@lo, kind: FK_Data_4
+v_mad_u32_u24 v0, g0@abs32@lo, v0, 12
+
+// GFX10: v_mad_u32_u24 v0, v0, g0@abs32@lo, 12   ; encoding: [0x00,0x00,0x43,0xd5,0x00,0xff,0x31,0x02,A,A,A,A]
+// GFX10-NEXT:                                    ;   fixup A - offset: 8, value: g0@abs32@lo, kind: FK_Data_4
+v_mad_u32_u24 v0, v0, g0@abs32@lo, 12
+
+// GFX10: v_mad_u32_u24 v0, v0, 12, g0@abs32@lo   ; encoding: [0x00,0x00,0x43,0xd5,0x00,0x19,0xfd,0x03,A,A,A,A]
+// GFX10-NEXT:                                    ;   fixup A - offset: 8, value: g0@abs32@lo, kind: FK_Data_4
+v_mad_u32_u24 v0, v0, 12, g0@abs32@lo
+
+; test vop2 operands
+
+// GFX10: v_add_nc_u32_e32 v0, g0@abs32@lo, v1    ; encoding: [0xff,0x02,0x00,0x4a,A,A,A,A]
+// GFX10-NEXT:                                    ;   fixup A - offset: 4, value: g0@abs32@lo, kind: FK_Data_4
+v_add_nc_u32 v0, g0@abs32@lo, v1
+
+// GFX10: v_add_nc_u32_e64 v0, v1, g0@abs32@lo    ; encoding: [0x00,0x00,0x25,0xd5,0x01,0xff,0x01,0x00,A,A,A,A]
+// GFX10-NEXT:                                    ;   fixup A - offset: 8, value: g0@abs32@lo, kind: FK_Data_4
+v_add_nc_u32 v0, v1, g0@abs32@lo
+
+// test vop1 operands
+// GFX10: v_not_b32_e32 v0, g0@abs32@lo           ; encoding: [0xff,0x6e,0x00,0x7e,A,A,A,A]
+// GFX10-NEXT:                                    ;   fixup A - offset: 4, value: g0@abs32@lo, kind: FK_Data_4
+v_not_b32 v0, g0@abs32@lo
+

[jayfoad]

LGTM with or without the suggestion - thanks!

[jayfoad]

Is this worth backporting to 19.x?

[arsenm]

Is this worth backporting to 19.x?

Sure