Verifier checks for whole wave funcs

Author: rovka

llvm/include/llvm/IR/CallingConv.h

@@ -297,8 +297,13 @@ namespace CallingConv {
 /// directly or indirectly via a call-like instruction.
 constexpr bool isCallableCC(CallingConv::ID CC) {
   switch (CC) {
+  // Called with special intrinsics:
+  // llvm.amdgcn.cs.chain
   case CallingConv::AMDGPU_CS_Chain:
   case CallingConv::AMDGPU_CS_ChainPreserve:
+  // llvm.amdgcn.call.whole.wave
+  case CallingConv::AMDGPU_Gfx_WholeWave:
+  // Hardware entry points:
   case CallingConv::AMDGPU_CS:
   case CallingConv::AMDGPU_ES:
   case CallingConv::AMDGPU_GS:

llvm/lib/IR/Verifier.cpp

@@ -2975,6 +2975,16 @@ void Verifier::visitFunction(const Function &F) {
           "perfect forwarding!",
           &F);
     break;
+  case CallingConv::AMDGPU_Gfx_WholeWave:
+    Check(F.arg_size() != 0 && F.arg_begin()->getType()->isIntegerTy(1),
+          "Calling convention requires first argument to be i1", &F);
+    Check(!F.arg_begin()->hasInRegAttr(),
+          "Calling convention requires first argument to not be inreg", &F);
+    Check(!F.isVarArg(),
+          "Calling convention does not support varargs or "
+          "perfect forwarding!",
+          &F);
+    break;
   }
 
   // Check that the argument values match the function type for this function...

llvm/test/Bitcode/compatibility.ll

@@ -564,6 +564,10 @@ declare riscv_vls_cc(32768) void @riscv_vls_cc_32768()
 ; CHECK: declare riscv_vls_cc(32768) void @riscv_vls_cc_32768()
 declare riscv_vls_cc(65536) void @riscv_vls_cc_65536()
 ; CHECK: declare riscv_vls_cc(65536) void @riscv_vls_cc_65536()
+declare cc124 void @f.cc124(i1)
+; CHECK: declare amdgpu_gfx_whole_wave void @f.cc124(i1)
+declare amdgpu_gfx_whole_wave void @f.amdgpu_gfx_whole_wave(i1)
+; CHECK: declare amdgpu_gfx_whole_wave void @f.amdgpu_gfx_whole_wave(i1)
 declare cc1023 void @f.cc1023()
 ; CHECK: declare cc1023 void @f.cc1023()
 

llvm/test/Verifier/amdgpu-cc.ll

@@ -217,3 +217,36 @@ define amdgpu_cs_chain_preserve void @preallocated_cc_amdgpu_cs_chain_preserve(p
 define amdgpu_cs_chain_preserve void @inalloca_cc_amdgpu_cs_chain_preserve(ptr inalloca(i32) %ptr) {
   ret void
 }
+
+; CHECK: Calling convention requires first argument to be i1
+; CHECK-NEXT: ptr @whole_wave_no_args
+define amdgpu_gfx_whole_wave void @whole_wave_no_args() {
+  ret void
+}
+
+; CHECK: Calling convention requires first argument to be i1
+; CHECK-NEXT: ptr @whole_wave_must_have_i1_active
+define amdgpu_gfx_whole_wave void @whole_wave_must_have_i1_active(i32 %x) {
+  ret void
+}
+
+; CHECK: Calling convention requires first argument to not be inreg
+; CHECK-NEXT: ptr @whole_wave_i1_active_inreg
+define amdgpu_gfx_whole_wave void @whole_wave_i1_active_inreg(i1 inreg %active) {
+  ret void
+}
+
+; CHECK: Calling convention does not support varargs
+; CHECK-NEXT: ptr @whole_wave_varargs
+define amdgpu_gfx_whole_wave void @whole_wave_varargs(i1 %active, i32 %x, ...) {
+  ret void
+}
+
+declare amdgpu_gfx_whole_wave void @whole_wave_callee(i1 %active)
+
+; CHECK: calling convention does not permit calls
+; CHECK-NEXT: call amdgpu_gfx_whole_wave void @whole_wave_callee(i1 true)
+define amdgpu_cs void @cant_call_whole_wave_func() {
+  call amdgpu_gfx_whole_wave void @whole_wave_callee(i1 true)
+  ret void
+}