triton/test/Conversion/triton_to_tritongpu.mlir

// RUN: triton-opt %s -convert-triton-to-tritongpu=num-warps=2 | FileCheck %s

func @ops() {
// CHECK: module attributes {"triton_gpu.num-warps" = 2 : i32} {{.*}}
  %a = arith.constant dense<1.00e+00> : tensor<128x32xf16>
  %b = arith.constant dense<2.00e+00> : tensor<32x128xf16>
  %c = arith.constant dense<3.00e+00> : tensor<128x128xf32>
  %0 = tt.dot %a, %b, %c {allowTF32 = true} : tensor<128x32xf16> * tensor<32x128xf16> -> tensor<128x128xf32>
  return
}