[Triton-MLIR]Add ptx vprintf support (#825)

Not know how to write unit test for this feature.

Co-authored-by: Yan Chunwei <yanchunwei@outlook.com>

include/triton/Dialect/Triton/IR/TritonOps.td

@@ -30,8 +30,8 @@ class TT_Op<string mnemonic, list<Trait> traits = []> :
 //   fptoui, fptosi, uitofp, sitofp,
 //   extf, tructf,
 //   extui, extsi, tructi
-def TT_IntToPtrOp : TT_Op<"int_to_ptr", [SameOperandsAndResultShape, 
-                                         SameOperandsAndResultEncoding, 
+def TT_IntToPtrOp : TT_Op<"int_to_ptr", [SameOperandsAndResultShape,
+                                         SameOperandsAndResultEncoding,
                                          NoSideEffect,
                                          /*DeclareOpInterfaceMethods<CastOpInterface>*/]> {
     let summary = "Cast int64 to pointer";
@@ -43,7 +43,7 @@ def TT_IntToPtrOp : TT_Op<"int_to_ptr", [SameOperandsAndResultShape,
     let assemblyFormat = "$from attr-dict `:` type($from) `->` type($result)";
 }
 
-def TT_PtrToIntOp : TT_Op<"ptr_to_int", [SameOperandsAndResultShape, 
+def TT_PtrToIntOp : TT_Op<"ptr_to_int", [SameOperandsAndResultShape,
                                          SameOperandsAndResultEncoding,
                                          NoSideEffect,
                                          /*DeclareOpInterfaceMethods<CastOpInterface>*/]> {
@@ -57,7 +57,7 @@ def TT_PtrToIntOp : TT_Op<"ptr_to_int", [SameOperandsAndResultShape,
 }
 
 // arith.bitcast doesn't support pointers
-def TT_BitcastOp : TT_Op<"bitcast", [SameOperandsAndResultShape, 
+def TT_BitcastOp : TT_Op<"bitcast", [SameOperandsAndResultShape,
                                      SameOperandsAndResultEncoding,
                                      NoSideEffect,
                                      /*DeclareOpInterfaceMethods<CastOpInterface>*/]> {
@@ -72,7 +72,7 @@ def TT_BitcastOp : TT_Op<"bitcast", [SameOperandsAndResultShape,
     // TODO: Add verifier
 }
 
-def TT_FpToFp : TT_Op<"fp_to_fp", [SameOperandsAndResultShape, 
+def TT_FpToFp : TT_Op<"fp_to_fp", [SameOperandsAndResultShape,
                                    SameOperandsAndResultEncoding,
                                    NoSideEffect,
                                    /*DeclareOpInterfaceMethods<CastOpInterface>*/]> {
@@ -99,7 +99,7 @@ def TT_FpToFp : TT_Op<"fp_to_fp", [SameOperandsAndResultShape,
 //
 
 def TT_AddPtrOp : TT_Op<"addptr",
-                     [NoSideEffect, 
+                     [NoSideEffect,
                      SameOperandsAndResultShape,
                      SameOperandsAndResultEncoding,
                       TypesMatchWith<"result type matches ptr type",
@@ -224,7 +224,7 @@ def TT_AtomicCASOp : TT_Op<"atomic_cas", [SameOperandsAndResultShape,
 //
 // Shape Manipulation Ops
 //
-def TT_SplatOp : TT_Op<"splat", [NoSideEffect, 
+def TT_SplatOp : TT_Op<"splat", [NoSideEffect,
                                  SameOperandsAndResultElementType]> {
     let summary = "splat";
 
@@ -237,8 +237,8 @@ def TT_SplatOp : TT_Op<"splat", [NoSideEffect,
     let hasFolder = 1;
 }
 
-def TT_ExpandDimsOp : TT_Op<"expand_dims", [NoSideEffect, 
-                                            DeclareOpInterfaceMethods<InferTypeOpInterface>, 
+def TT_ExpandDimsOp : TT_Op<"expand_dims", [NoSideEffect,
+                                            DeclareOpInterfaceMethods<InferTypeOpInterface>,
                                             SameOperandsAndResultElementType]> {
     let summary = "expand_dims";
 
@@ -249,7 +249,7 @@ def TT_ExpandDimsOp : TT_Op<"expand_dims", [NoSideEffect,
     let assemblyFormat = "$src attr-dict `:` functional-type(operands, results)";
 }
 
-def TT_ViewOp : TT_Op<"view", [NoSideEffect, 
+def TT_ViewOp : TT_Op<"view", [NoSideEffect,
                                SameOperandsAndResultElementType]> {
     let summary = "view";
 
@@ -261,7 +261,7 @@ def TT_ViewOp : TT_Op<"view", [NoSideEffect,
 
 }
 
-def TT_BroadcastOp : TT_Op<"broadcast", [NoSideEffect, 
+def TT_BroadcastOp : TT_Op<"broadcast", [NoSideEffect,
                                          SameOperandsAndResultElementType]> {
     let summary = "broadcast. No left-padding as of now.";
 
@@ -274,7 +274,7 @@ def TT_BroadcastOp : TT_Op<"broadcast", [NoSideEffect,
     let hasFolder = 1;
 }
 
-def TT_CatOp : TT_Op<"cat", [NoSideEffect, 
+def TT_CatOp : TT_Op<"cat", [NoSideEffect,
                              SameOperandsAndResultElementType]> {
     let summary = "concatenate 2 tensors";
 
@@ -307,7 +307,7 @@ def TT_GetNumProgramsOp : TT_Op<"get_num_programs", [NoSideEffect]> {
 //
 // Dot Op
 //
-def TT_DotOp : TT_Op<"dot", [NoSideEffect, 
+def TT_DotOp : TT_Op<"dot", [NoSideEffect,
                              DeclareOpInterfaceMethods<InferTypeOpInterface>,
                              TypesMatchWith<"result's type matches accumulator's type",
                                             "d", "c", "$_self">]> {
@@ -385,4 +385,20 @@ def TT_MakeRangeOp : TT_Op<"make_range", [NoSideEffect]> {
     let assemblyFormat = "attr-dict `:` type($result)";
 }
 
+//
+// Make PrintfOp
+//
+def TT_PrintfOp : TT_Op<"printf", [MemoryEffects<[MemWrite]>]>,
+  Arguments<(ins StrAttr:$prefix,
+                Variadic<AnyTypeOf<[TT_Type]>>:$args)> {
+  let summary = "Device-side printf, as in CUDA for debugging";
+  let description = [{
+    `tt.printf` takes a literal string prefix and an arbitrary number of scalar or tensor arguments that should be printed.
+    format are generated automatically from the arguments.
+  }];
+  let assemblyFormat = [{
+    $prefix attr-dict ($args^ `:` type($args))?
+  }];
+}
+
 #endif // Triton_OPS