[RISCV] Add support for inline asm constraint vd

[tclin914]

It constrains vector registers excluding v0. Refer to https://gcc.gnu.org/onlinedocs/gcc/Machine-Constraints.html RISC-V part.

This patch also adds a testcase for constraints vr, vd and vm.

[llvmbot]

@llvm/pr-subscribers-clang

@llvm/pr-subscribers-backend-risc-v

Author: Jim Lin (tclin914)

Changes

It constrains vector registers excluding v0.

This patch also adds a testcase for constraints vr, vd and vm.

---

Full diff: https://github.com/llvm/llvm-project/pull/111653.diff

2 Files Affected:

• (modified) llvm/lib/Target/RISCV/RISCVISelLowering.cpp (+14-1)
• (added) llvm/test/CodeGen/RISCV/inline-asm-v-constraint.ll (+66)

diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 463887b8b55e61..0c8fd324edd1cf 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -20340,7 +20340,7 @@ RISCVTargetLowering::getConstraintType(StringRef Constraint) const {
       return C_Other;
     }
   } else {
-    if (Constraint == "vr" || Constraint == "vm")
+    if (Constraint == "vr" || Constraint == "vd" || Constraint == "vm")
       return C_RegisterClass;
   }
   return TargetLowering::getConstraintType(Constraint);
@@ -20388,6 +20388,19 @@ RISCVTargetLowering::getRegForInlineAsmConstraint(const TargetRegisterInfo *TRI,
       if (TRI->isTypeLegalForClass(*RC, VT.SimpleTy))
         return std::make_pair(0U, RC);
     }
+  } else if (Constraint == "vd") {
+    for (const auto *RC :
+         {&RISCV::VRNoV0RegClass, &RISCV::VRM2NoV0RegClass,
+          &RISCV::VRM4NoV0RegClass, &RISCV::VRM8NoV0RegClass,
+          &RISCV::VRN2M1NoV0RegClass, &RISCV::VRN3M1NoV0RegClass,
+          &RISCV::VRN4M1NoV0RegClass, &RISCV::VRN5M1NoV0RegClass,
+          &RISCV::VRN6M1NoV0RegClass, &RISCV::VRN7M1NoV0RegClass,
+          &RISCV::VRN8M1NoV0RegClass, &RISCV::VRN2M2NoV0RegClass,
+          &RISCV::VRN3M2NoV0RegClass, &RISCV::VRN4M2NoV0RegClass,
+          &RISCV::VRN2M4NoV0RegClass}) {
+      if (TRI->isTypeLegalForClass(*RC, VT.SimpleTy))
+        return std::make_pair(0U, RC);
+    }
   } else if (Constraint == "vm") {
     if (TRI->isTypeLegalForClass(RISCV::VMV0RegClass, VT.SimpleTy))
       return std::make_pair(0U, &RISCV::VMV0RegClass);
diff --git a/llvm/test/CodeGen/RISCV/inline-asm-v-constraint.ll b/llvm/test/CodeGen/RISCV/inline-asm-v-constraint.ll
new file mode 100644
index 00000000000000..c04e4fea7b2c29
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/inline-asm-v-constraint.ll
@@ -0,0 +1,66 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -mattr=+v -verify-machineinstrs < %s \
+; RUN:   | FileCheck -check-prefix=RV32I %s
+; RUN: llc -mtriple=riscv64 -mattr=+v -verify-machineinstrs < %s \
+; RUN:   | FileCheck -check-prefix=RV64I %s
+
+define <vscale x 1 x i8> @constraint_vr(<vscale x 1 x i8> %0, <vscale x 1 x i8> %1) nounwind {
+; RV32I-LABEL: constraint_vr:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    #APP
+; RV32I-NEXT:    vadd.vv v8, v8, v9
+; RV32I-NEXT:    #NO_APP
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: constraint_vr:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    #APP
+; RV64I-NEXT:    vadd.vv v8, v8, v9
+; RV64I-NEXT:    #NO_APP
+; RV64I-NEXT:    ret
+  %a = tail call <vscale x 1 x i8> asm "vadd.vv $0, $1, $2", "=^vr,^vr,^vr"(
+    <vscale x 1 x i8> %0, <vscale x 1 x i8> %1)
+  ret <vscale x 1 x i8> %a
+}
+
+define <vscale x 1 x i8> @constraint_vd(<vscale x 1 x i8> %0, <vscale x 1 x i8> %1) nounwind {
+; RV32I-LABEL: constraint_vd:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    #APP
+; RV32I-NEXT:    vadd.vv v8, v8, v9
+; RV32I-NEXT:    #NO_APP
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: constraint_vd:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    #APP
+; RV64I-NEXT:    vadd.vv v8, v8, v9
+; RV64I-NEXT:    #NO_APP
+; RV64I-NEXT:    ret
+  %a = tail call <vscale x 1 x i8> asm "vadd.vv $0, $1, $2", "=^vd,^vr,^vr"(
+    <vscale x 1 x i8> %0, <vscale x 1 x i8> %1)
+  ret <vscale x 1 x i8> %a
+}
+
+define <vscale x 1 x i1> @constraint_vm(<vscale x 1 x i1> %0, <vscale x 1 x i1> %1) nounwind {
+; RV32I-LABEL: constraint_vm:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    vmv1r.v v9, v0
+; RV32I-NEXT:    vmv1r.v v0, v8
+; RV32I-NEXT:    #APP
+; RV32I-NEXT:    vadd.vv v0, v9, v0
+; RV32I-NEXT:    #NO_APP
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: constraint_vm:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    vmv1r.v v9, v0
+; RV64I-NEXT:    vmv1r.v v0, v8
+; RV64I-NEXT:    #APP
+; RV64I-NEXT:    vadd.vv v0, v9, v0
+; RV64I-NEXT:    #NO_APP
+; RV64I-NEXT:    ret
+  %a = tail call <vscale x 1 x i1> asm "vadd.vv $0, $1, $2", "=^vr,^vr,^vm"(
+    <vscale x 1 x i1> %0, <vscale x 1 x i1> %1)
+  ret <vscale x 1 x i1> %a
+}

[lenary]

I think you've missed making the relevant clang changes to RISCVTargetInfo::validateAsmConstraint to support vd.

I don't understand why this constraint is needed but it's good to stay in line with GCC.

[lenary]

LGTM

[topperc]

This needs to be tested in clang/test/CodeGen/RISCV/riscv-inline-asm-rvv.c

[topperc]

Nevermind. I guess I'm blind.

[lenary]

@topperc is this test not enough?

[topperc]

weird how did i not see that was here. Sorry

[topperc]

LGTM