AMDGPU: add __builtin_amdgcn_update_dpp

[yaxunl]

Summary

Emit llvm.amdgcn.update.dpp for both __builtin_amdgcn_mov_dpp and
__builtin_amdgcn_update_dpp. The second last argument to
llvm.amdgcn.update.dpp will be undef for __builtin_amdgcn_mov_dpp.

[arsenm]

lib/CodeGen/CGBuiltin.cpp

11313–11315

The only difference between this and mov_dpp is the argument count and the intrinsic ID, so you can combine the cases

[b-sumner]

lib/CodeGen/CGBuiltin.cpp

11313–11315

We should really drop mov_dpp. It will be easier to do so if we keep the cases separate.

[arsenm]

lib/CodeGen/CGBuiltin.cpp

11313–11315

We could also emit the mov_dpp with the new intrinsic, using I think undef for the extra operand?

[b-sumner]

Ping. There's quite a bit of interest in getting this exposed by clang.

[yaxunl]

emit llvm.amdgcn.update.dpp for __builtin_amdgcn_mov_dpp.

[arsenm]

LGTM

[yaxunl]

Brian checked the extra argument for dpp mov should be the first one. so mov_dpp(x,...) --> update_dpp(undef, x, ...). I will fix that when committing.

include/clang/Basic/BuiltinsAMDGPU.def

 TARGET_BUILTIN(__builtin_amdgcn_cosh, "hh", "nc", "16-bit-insts")
 TARGET_BUILTIN(__builtin_amdgcn_ldexph, "hhi", "nc", "16-bit-insts")
 TARGET_BUILTIN(__builtin_amdgcn_frexp_manth, "hh", "nc", "16-bit-insts")
 TARGET_BUILTIN(__builtin_amdgcn_frexp_exph, "sh", "nc", "16-bit-insts")
 TARGET_BUILTIN(__builtin_amdgcn_fracth, "hh", "nc", "16-bit-insts")
 TARGET_BUILTIN(__builtin_amdgcn_classh, "bhi", "nc", "16-bit-insts")
 TARGET_BUILTIN(__builtin_amdgcn_s_memrealtime, "LUi", "n", "s-memrealtime")
 TARGET_BUILTIN(__builtin_amdgcn_mov_dpp, "iiIiIiIiIb", "nc", "dpp")
+TARGET_BUILTIN(__builtin_amdgcn_update_dpp, "iiiIiIiIiIb", "nc", "dpp")
 TARGET_BUILTIN(__builtin_amdgcn_s_dcache_wb, "v", "n", "vi-insts")
 
 //===----------------------------------------------------------------------===//
 // GFX9+ only builtins.
 //===----------------------------------------------------------------------===//
 
 TARGET_BUILTIN(__builtin_amdgcn_fmed3h, "hhhh", "nc", "gfx9-insts")
 

lib/CodeGen/CGBuiltin.cpp

 
 Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
                                               const CallExpr *E) {
   switch (BuiltinID) {
   case AMDGPU::BI__builtin_amdgcn_div_scale:
   case AMDGPU::BI__builtin_amdgcn_div_scalef: {
     // Translate from the intrinsics's struct return to the builtin's out
     // argument.
 
     Address FlagOutPtr = EmitPointerWithAlignment(E->getArg(3));
 
     llvm::Value *X = EmitScalarExpr(E->getArg(0));
     llvm::Value *Y = EmitScalarExpr(E->getArg(1));
     llvm::Value *Z = EmitScalarExpr(E->getArg(2));
 
     llvm::Value *Callee = CGM.getIntrinsic(Intrinsic::amdgcn_div_scale,
                                            X->getType());
 
     llvm::Value *Tmp = Builder.CreateCall(Callee, {X, Y, Z});
case AMDGPU::BI__builtin_amdgcn_div_fmasf: {
     llvm::Value *F = CGM.getIntrinsic(Intrinsic::amdgcn_div_fmas,
                                       Src0->getType());
     llvm::Value *Src3ToBool = Builder.CreateIsNotNull(Src3);
     return Builder.CreateCall(F, {Src0, Src1, Src2, Src3ToBool});
   }
 
   case AMDGPU::BI__builtin_amdgcn_ds_swizzle:
     return emitBinaryBuiltin(*this, E, Intrinsic::amdgcn_ds_swizzle);
-  case AMDGPU::BI__builtin_amdgcn_mov_dpp: {
-    llvm::SmallVector<llvm::Value *, 5> Args;
-    for (unsigned I = 0; I != 5; ++I)
+  case AMDGPU::BI__builtin_amdgcn_mov_dpp:
+  case AMDGPU::BI__builtin_amdgcn_update_dpp: {
+    llvm::SmallVector<llvm::Value *, 6> Args;
+    for (unsigned I = 0; I != E->getNumArgs(); ++I)
       Args.push_back(EmitScalarExpr(E->getArg(I)));
-    Value *F = CGM.getIntrinsic(Intrinsic::amdgcn_mov_dpp,
-                                    Args[0]->getType());
+    assert(Args.size() == 5 || Args.size() == 6);
+    if (Args.size() == 5)
+      Args.insert(Args.begin(), llvm::UndefValue::get(Args[0]->getType()));
+    Value *F =
+        CGM.getIntrinsic(Intrinsic::amdgcn_update_dpp, Args[0]->getType());
     return Builder.CreateCall(F, Args);
   }
   case AMDGPU::BI__builtin_amdgcn_div_fixup:
   case AMDGPU::BI__builtin_amdgcn_div_fixupf:
   case AMDGPU::BI__builtin_amdgcn_div_fixuph:
     return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_div_fixup);
   case AMDGPU::BI__builtin_amdgcn_trig_preop:
   case AMDGPU::BI__builtin_amdgcn_trig_preopf:

test/CodeGenOpenCL/builtins-amdgcn-vi.cl

 // CHECK-LABEL: @test_s_dcache_wb()
 // CHECK: call void @llvm.amdgcn.s.dcache.wb()
 void test_s_dcache_wb()
 {
   __builtin_amdgcn_s_dcache_wb();
 }
 
 // CHECK-LABEL: @test_mov_dpp
-// CHECK: call i32 @llvm.amdgcn.mov.dpp.i32(i32 %src, i32 0, i32 0, i32 0, i1 false)
+// CHECK: call i32 @llvm.amdgcn.update.dpp.i32(i32 undef, i32 %src, i32 0, i32 0, i32 0, i1 false)
 void test_mov_dpp(global int* out, int src)
 {
   *out = __builtin_amdgcn_mov_dpp(src, 0, 0, 0, false);
 }
 
+// CHECK-LABEL: @test_update_dpp
+// CHECK: call i32 @llvm.amdgcn.update.dpp.i32(i32 %arg1, i32 %arg2, i32 0, i32 0, i32 0, i1 false)
+void test_update_dpp(global int* out, int arg1, int arg2)
+{
+  *out = __builtin_amdgcn_update_dpp(arg1, arg2, 0, 0, 0, false);
+}
+
 // CHECK-LABEL: @test_ds_fadd
 // CHECK: call float @llvm.amdgcn.ds.fadd(float addrspace(3)* %out, float %src, i32 0, i32 0, i1 false)
 void test_ds_faddf(local float *out, float src) {
   *out = __builtin_amdgcn_ds_faddf(out, src, 0, 0, false);
 }
 
 // CHECK-LABEL: @test_ds_fmin
 // CHECK: call float @llvm.amdgcn.ds.fmin(float addrspace(3)* %out, float %src, i32 0, i32 0, i1 false)

test/SemaOpenCL/builtins-amdgcn-error.cl

 void test_mov_dpp2(global int* out, int a, int b, int c, int d, bool e)
 {
   *out = __builtin_amdgcn_mov_dpp(a, b, 0, 0, false); // expected-error {{argument to '__builtin_amdgcn_mov_dpp' must be a constant integer}}
   *out = __builtin_amdgcn_mov_dpp(a, 0, c, 0, false); // expected-error {{argument to '__builtin_amdgcn_mov_dpp' must be a constant integer}}
   *out = __builtin_amdgcn_mov_dpp(a, 0, 0, d, false); // expected-error {{argument to '__builtin_amdgcn_mov_dpp' must be a constant integer}}
   *out = __builtin_amdgcn_mov_dpp(a, 0, 0, 0, e); // expected-error {{argument to '__builtin_amdgcn_mov_dpp' must be a constant integer}}
 }
 
+void test_update_dpp2(global int* out, int a, int b, int c, int d, int e, bool f)
+{
+  *out = __builtin_amdgcn_update_dpp(a, b, 0, 0, 0, false);
+  *out = __builtin_amdgcn_update_dpp(a, 0, c, 0, 0, false); // expected-error {{argument to '__builtin_amdgcn_update_dpp' must be a constant integer}}
+  *out = __builtin_amdgcn_update_dpp(a, 0, 0, d, 0, false); // expected-error {{argument to '__builtin_amdgcn_update_dpp' must be a constant integer}}
+  *out = __builtin_amdgcn_update_dpp(a, 0, 0, 0, e, false); // expected-error {{argument to '__builtin_amdgcn_update_dpp' must be a constant integer}}
+  *out = __builtin_amdgcn_update_dpp(a, 0, 0, 0, 0, f); // expected-error {{argument to '__builtin_amdgcn_update_dpp' must be a constant integer}}
+}
+
 void test_ds_faddf(local float *out, float src, int a) {
   *out = __builtin_amdgcn_ds_faddf(out, src, a, 0, false); // expected-error {{argument to '__builtin_amdgcn_ds_faddf' must be a constant integer}}
   *out = __builtin_amdgcn_ds_faddf(out, src, 0, a, false); // expected-error {{argument to '__builtin_amdgcn_ds_faddf' must be a constant integer}}
   *out = __builtin_amdgcn_ds_faddf(out, src, 0, 0, a); // expected-error {{argument to '__builtin_amdgcn_ds_faddf' must be a constant integer}}
 }
 
 void test_ds_fminf(local float *out, float src, int a) {
   *out = __builtin_amdgcn_ds_fminf(out, src, a, 0, false); // expected-error {{argument to '__builtin_amdgcn_ds_fminf' must be a constant integer}}