Created
March 27, 2024 03:09
-
-
Save antiagainst/acb8c2b7fdc9d538210d42b0c62858cd to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
#loc = loc(unknown) | |
module attributes {"triton_gpu.compute-capability" = 90 : i32, "triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 4 : i32, triton_gpu.shared = 0 : i32, "triton_gpu.threads-per-warp" = 64 : i32} { | |
llvm.func @__ockl_printf_append_args(i64, i32, i64, i64, i64, i64, i64, i64, i64, i32) -> i64 loc(#loc) | |
llvm.func @__ockl_printf_append_string_n(i64, !llvm.ptr, i64, i32) -> i64 loc(#loc) | |
llvm.func @__ockl_printf_begin(i64) -> i64 loc(#loc) | |
llvm.mlir.global internal constant @printfFormat_0("pid (%u, %u, %u) idx (%3u) x: %u\0A\00") {addr_space = 0 : i32} loc(#loc) | |
llvm.mlir.global internal constant @printfPrefix_0(" x: ") {addr_space = 0 : i32} loc(#loc) | |
llvm.mlir.global external @global_smem() {addr_space = 3 : i32, alignment = 16 : i64} : !llvm.array<0 x i8> loc(#loc) | |
llvm.func @kernel_device_print_0d1d(%arg0: !llvm.ptr<1> {tt.divisibility = 16 : i32} loc(unknown), %arg1: !llvm.ptr<1> {tt.divisibility = 16 : i32} loc(unknown)) attributes {noinline = false, nvvm.kernel = 1 : ui1, nvvm.maxntid = array<i32: 128>} { | |
%0 = llvm.mlir.constant(5 : i32) : i32 loc(#loc) | |
%1 = llvm.mlir.constant(34 : i64) : i64 loc(#loc) | |
%2 = llvm.mlir.constant(0 : i64) : i64 loc(#loc) | |
%3 = llvm.mlir.constant(true) : i1 loc(#loc) | |
%4 = llvm.mlir.constant(0 : i32) : i32 loc(#loc) | |
%5 = llvm.mlir.constant(1 : i32) : i32 loc(#loc) | |
%6 = llvm.mlir.constant(128 : i32) : i32 loc(#loc) | |
%7 = llvm.mlir.constant(2 : i32) : i32 loc(#loc) | |
%8 = llvm.mlir.constant(4 : i32) : i32 loc(#loc) | |
%9 = llvm.mlir.constant(64 : i32) : i32 loc(#loc) | |
%10 = llvm.mlir.constant(0 : index) : i32 loc(#loc) | |
%11 = rocdl.workitem.id.x : i32 loc(#loc) | |
%12 = llvm.urem %11, %9 : i32 loc(#loc) | |
%13 = llvm.udiv %11, %9 : i32 loc(#loc) | |
%14 = llvm.urem %13, %8 : i32 loc(#loc) | |
%15 = llvm.urem %12, %9 : i32 loc(#loc) | |
%16 = llvm.urem %14, %7 : i32 loc(#loc) | |
%17 = llvm.urem %15, %6 : i32 loc(#loc) | |
%18 = llvm.mul %16, %9 : i32 loc(#loc) | |
%19 = llvm.add %17, %18 : i32 loc(#loc) | |
%20 = llvm.mul %19, %5 : i32 loc(#loc) | |
%21 = llvm.urem %4, %5 : i32 loc(#loc) | |
%22 = llvm.urem %21, %5 : i32 loc(#loc) | |
%23 = llvm.mul %22, %6 : i32 loc(#loc) | |
%24 = llvm.add %20, %23 : i32 loc(#loc) | |
%25 = llvm.add %24, %4 : i32 loc(#loc) | |
%26 = llvm.add %25, %10 : i32 loc(#loc) | |
%27 = llvm.getelementptr %arg0[%26] : (!llvm.ptr<1>, i32) -> !llvm.ptr<1>, i32 loc(#loc) | |
%28 = llvm.addrspacecast %27 : !llvm.ptr<1> to !llvm.ptr loc(#loc) | |
llvm.cond_br %3, ^bb1, ^bb2 loc(#loc) | |
^bb1: // pred: ^bb0 | |
%29 = llvm.load %28 : !llvm.ptr -> i32 loc(#loc) | |
llvm.br ^bb3(%29 : i32) loc(#loc) | |
^bb2: // pred: ^bb0 | |
llvm.br ^bb3(%4 : i32) loc(#loc) | |
^bb3(%30: i32 loc(unknown)): // 2 preds: ^bb1, ^bb2 | |
llvm.br ^bb4 loc(#loc) | |
^bb4: // pred: ^bb3 | |
%31 = llvm.bitcast %30 : i32 to vector<1xi32> loc(#loc) | |
%32 = llvm.extractelement %31[%10 : i32] : vector<1xi32> loc(#loc) | |
%33 = rocdl.workgroup.id.x : i32 loc(#loc) | |
%34 = rocdl.workgroup.id.y : i32 loc(#loc) | |
%35 = rocdl.workgroup.id.z : i32 loc(#loc) | |
%36 = llvm.mlir.addressof @printfFormat_0 : !llvm.ptr loc(#loc) | |
%37 = llvm.call @__ockl_printf_begin(%2) : (i64) -> i64 loc(#loc) | |
%38 = llvm.call @__ockl_printf_append_string_n(%37, %36, %1, %4) : (i64, !llvm.ptr, i64, i32) -> i64 loc(#loc) | |
%39 = llvm.sext %33 : i32 to i64 loc(#loc) | |
%40 = llvm.sext %34 : i32 to i64 loc(#loc) | |
%41 = llvm.sext %35 : i32 to i64 loc(#loc) | |
%42 = llvm.sext %25 : i32 to i64 loc(#loc) | |
%43 = llvm.sext %32 : i32 to i64 loc(#loc) | |
%44 = llvm.call @__ockl_printf_append_args(%38, %0, %39, %40, %41, %42, %43, %2, %2, %5) : (i64, i32, i64, i64, i64, i64, i64, i64, i64, i32) -> i64 loc(#loc) | |
%45 = llvm.getelementptr %arg1[%26] : (!llvm.ptr<1>, i32) -> !llvm.ptr<1>, i32 loc(#loc) | |
%46 = llvm.mul %14, %9 : i32 loc(#loc) | |
%47 = llvm.add %46, %15 : i32 loc(#loc) | |
%48 = llvm.mul %47, %5 : i32 loc(#loc) | |
%49 = llvm.icmp "slt" %48, %6 : i32 loc(#loc) | |
%50 = llvm.and %3, %49 : i1 loc(#loc) | |
%51 = llvm.mlir.undef : vector<1xi32> loc(#loc) | |
%52 = llvm.insertelement %32, %51[%4 : i32] : vector<1xi32> loc(#loc) | |
%53 = llvm.bitcast %52 : vector<1xi32> to i32 loc(#loc) | |
llvm.cond_br %50, ^bb5, ^bb6 loc(#loc) | |
^bb5: // pred: ^bb4 | |
llvm.store %53, %45 : i32, !llvm.ptr<1> loc(#loc) | |
llvm.br ^bb6 loc(#loc) | |
^bb6: // 2 preds: ^bb4, ^bb5 | |
llvm.return loc(#loc) | |
} loc(#loc) | |
} loc(#loc) |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
module attributes {gpu.container_module} { | |
gpu.module @kernels [#rocdl.target<chip = "gfx942">] { | |
llvm.mlir.global internal constant @printfFormat_0("Hello from %d\0A\00") {addr_space = 0 : i32} | |
llvm.func @__ockl_printf_append_string_n(i64, !llvm.ptr, i64, i32) -> i64 | |
llvm.func @__ockl_printf_append_args(i64, i32, i64, i64, i64, i64, i64, i64, i64, i32) -> i64 | |
llvm.func @__ockl_printf_begin(i64) -> i64 | |
llvm.func @hello() attributes {gpu.kernel, rocdl.kernel} { | |
%0 = rocdl.workitem.id.x : i32 | |
%1 = llvm.mlir.constant(0 : i64) : i64 | |
%2 = llvm.call @__ockl_printf_begin(%1) : (i64) -> i64 | |
%3 = llvm.mlir.addressof @printfFormat_0 : !llvm.ptr | |
%4 = llvm.getelementptr %3[0, 0] : (!llvm.ptr) -> !llvm.ptr, !llvm.array<15 x i8> | |
%5 = llvm.mlir.constant(15 : i64) : i64 | |
%6 = llvm.mlir.constant(1 : i32) : i32 | |
%7 = llvm.mlir.constant(0 : i32) : i32 | |
%8 = llvm.call @__ockl_printf_append_string_n(%2, %4, %5, %7) : (i64, !llvm.ptr, i64, i32) -> i64 | |
%9 = llvm.mlir.constant(1 : i32) : i32 | |
%10 = llvm.zext %0 : i32 to i64 | |
%11 = llvm.call @__ockl_printf_append_args(%8, %9, %10, %1, %1, %1, %1, %1, %1, %6) : (i64, i32, i64, i64, i64, i64, i64, i64, i64, i32) -> i64 | |
llvm.return | |
} | |
} | |
llvm.func @main() { | |
%0 = llvm.mlir.constant(2 : index) : i64 | |
%1 = llvm.mlir.constant(1 : index) : i64 | |
gpu.launch_func @kernels::@hello blocks in (%1, %1, %1) threads in (%0, %1, %1) : i64 | |
llvm.return | |
} | |
} |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment