Update src/KernelAbstractions.jl

Co-authored-by: Simon Byrne <[email protected]>

Author: charleskawczynski

src/KernelAbstractions.jl

@@ -238,30 +238,25 @@ macro print(items...)
     end
 end
 
-@generated function promote_c_argument(arg)
-    # > When a function with a variable-length argument list is called, the variable
-    # > arguments are passed using C's old ``default argument promotions.'' These say that
-    # > types char and short int are automatically promoted to int, and type float is
-    # > automatically promoted to double. Therefore, varargs functions will never receive
-    # > arguments of type char, short int, or float.
-
-    if arg == Cchar || arg == Cshort
-        return :(Cint(arg))
-    elseif arg == Cfloat
-        return :(Cdouble(arg))
-    else
-        return :(arg)
-    end
-end
+# When a function with a variable-length argument list is called, the variable
+# arguments are passed using C's old ``default argument promotions.'' These say that
+# types char and short int are automatically promoted to int, and type float is
+# automatically promoted to double. Therefore, varargs functions will never receive
+# arguments of type char, short int, or float.
+
+promote_c_argument(arg) = arg
+promote_c_argument(arg::Cfloat) = Cdouble(arg)
+promote_c_argument(arg::Cchar) = Cint(arg)
+promote_c_argument(arg::Cshort) = Cint(arg)
 
 """
     @printf(fmt::String, args...)
 
-This is a unified formatted print statement.
+This is a unified formatted printf statement.
 
 # Platform differences
   - `GPU`: This will reorganize the items to print via @cuprintf
-  - `CPU`: This will call `print(items...)`
+  - `CPU`: This will call `sprintf(fmt, items...)`
 """
 macro printf(fmt::String, args...)
     fmt_val = Val(Symbol(fmt))
@@ -551,9 +546,7 @@ end
     end
     sfmt = String(fmt)
     quote
-        # @sprintf($sfmt, $(args...))
-        @print(@sprintf($sfmt, $(args...)))
-        # @print("test")
+        Printf.@printf($sfmt, $(args...))
     end
 end
 

src/backends/cuda.jl

@@ -320,7 +320,11 @@ end
 end
 
 @inline function Cassette.overdub(ctx::CUDACtx, ::typeof(__printf), fmt, args...)
-    CUDA._cuprintf(fmt, args...)
+    CUDA._cuprintf(Val(fmt), args...)
+end
+
+@inline function Cassette.overdub(ctx::CUDACtx, ::typeof(__printf), ::Val{fmt}, args...) where fmt
+    CUDA._cuprintf(Val(fmt), args...)
 end
 
 ###

test/print_test.jl

@@ -6,6 +6,7 @@ if has_cuda_gpu()
 end
 
 struct Foo{A,B} end
+get_name(::Type{T}) where T<:Foo = "Foo"
 
 @kernel function kernel_print()
   I = @index(Global)
@@ -14,7 +15,11 @@ end
 
 @kernel function kernel_printf()
   I = @index(Global)
-  @printf("Hello printf %s thread %d! type = %s.\n", "from", I, nameof(Foo))
+  # @printf("Hello printf %s thread %d! type = %s.\n", "from", I, nameof(Foo))
+  # @print("Hello printf from thread ", I, "!\n")
+  # @printf("Hello printf %s thread %d! type = %s.\n", "from", I, string(nameof(Foo)))
+  @printf("Hello printf %s thread %d! type = %s.\n", "from", I, "Foo")
+  @printf("Hello printf %s thread %d! type = %s.\n", "from", I, get_name(Foo))
 end
 
 function test_print(backend)