Add files using upload-large-folder tool

.triton/dump/99f0a4c15ca0aab38ccdae6c765f7333/triton_.ttir

@@ -0,0 +1,67 @@
+module {
+  tt.func public @triton__0d1d2d3d4d5d6de7de(%arg0: !tt.ptr<f32, 1> {tt.divisibility = 16 : i32}, %arg1: !tt.ptr<bf16, 1> {tt.divisibility = 16 : i32}, %arg2: !tt.ptr<bf16, 1> {tt.divisibility = 16 : i32}, %arg3: !tt.ptr<bf16, 1> {tt.divisibility = 16 : i32}, %arg4: !tt.ptr<f32, 1> {tt.divisibility = 16 : i32}, %arg5: !tt.ptr<bf16, 1> {tt.divisibility = 16 : i32}, %arg6: i32 {tt.divisibility = 16 : i32, tt.max_divisibility = 16 : i32}, %arg7: i32 {tt.divisibility = 16 : i32, tt.max_divisibility = 16 : i32}) attributes {noinline = false} {
+    %c256_i32 = arith.constant 256 : i32
+    %cst = arith.constant dense<0.000000e+00> : tensor<256xbf16>
+    %cst_0 = arith.constant 0.000000e+00 : f32
+    %cst_1 = arith.constant 2.560000e+02 : f32
+    %cst_2 = arith.constant 9.99999974E-6 : f32
+    %cst_3 = arith.constant dense<0.000000e+00> : tensor<256xf32>
+    %cst_4 = arith.constant dense<256> : tensor<256xi32>
+    %0 = tt.get_program_id x : i32
+    %1 = tt.make_range {end = 256 : i32, start = 0 : i32} : tensor<256xi32>
+    %2 = arith.cmpi slt, %1, %cst_4 : tensor<256xi32>
+    %3 = arith.muli %0, %c256_i32 : i32
+    %4 = tt.splat %3 : (i32) -> tensor<256xi32>
+    %5 = arith.addi %1, %4 : tensor<256xi32>
+    %6 = tt.splat %arg0 : (!tt.ptr<f32, 1>) -> tensor<256x!tt.ptr<f32, 1>>
+    %7 = tt.addptr %6, %5 : tensor<256x!tt.ptr<f32, 1>>, tensor<256xi32>
+    %8 = tt.load %7, %2, %cst_3 {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<256xf32>
+    %9 = tt.splat %arg1 : (!tt.ptr<bf16, 1>) -> tensor<256x!tt.ptr<bf16, 1>>
+    %10 = tt.addptr %9, %5 : tensor<256x!tt.ptr<bf16, 1>>, tensor<256xi32>
+    %11 = tt.load %10, %2, %cst {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<256xbf16>
+    %12 = arith.extf %11 : tensor<256xbf16> to tensor<256xf32>
+    %13 = tt.splat %arg2 : (!tt.ptr<bf16, 1>) -> tensor<256x!tt.ptr<bf16, 1>>
+    %14 = tt.addptr %13, %5 : tensor<256x!tt.ptr<bf16, 1>>, tensor<256xi32>
+    %15 = tt.load %14, %2, %cst {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<256xbf16>
+    %16 = arith.extf %15 : tensor<256xbf16> to tensor<256xf32>
+    %17 = tt.splat %arg3 : (!tt.ptr<bf16, 1>) -> tensor<256x!tt.ptr<bf16, 1>>
+    %18 = tt.addptr %17, %5 : tensor<256x!tt.ptr<bf16, 1>>, tensor<256xi32>
+    %19 = tt.load %18, %2, %cst {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<256xbf16>
+    %20 = arith.extf %19 : tensor<256xbf16> to tensor<256xf32>
+    %21 = tt.splat %arg4 : (!tt.ptr<f32, 1>) -> tensor<256x!tt.ptr<f32, 1>>
+    %22 = tt.addptr %21, %1 : tensor<256x!tt.ptr<f32, 1>>, tensor<256xi32>
+    %23 = tt.load %22, %2, %cst_3 {cache = 1 : i32, evict = 3 : i32, isVolatile = false} : tensor<256xf32>
+    %24 = arith.addf %8, %12 : tensor<256xf32>
+    %25 = arith.addf %24, %16 : tensor<256xf32>
+    %26 = arith.addf %25, %20 : tensor<256xf32>
+    %27 = arith.select %2, %26, %cst_3 : tensor<256xi1>, tensor<256xf32>
+    %28 = "tt.reduce"(%27) <{axis = 0 : i32}> ({
+    ^bb0(%arg8: f32, %arg9: f32):
+      %46 = arith.addf %arg8, %arg9 : f32
+      tt.reduce.return %46 : f32
+    }) : (tensor<256xf32>) -> f32
+    %29 = arith.addf %28, %cst_0 : f32
+    %30 = arith.divf %29, %cst_1 : f32
+    %31 = tt.splat %30 : (f32) -> tensor<256xf32>
+    %32 = arith.subf %26, %31 : tensor<256xf32>
+    %33 = arith.mulf %32, %32 : tensor<256xf32>
+    %34 = arith.select %2, %33, %cst_3 : tensor<256xi1>, tensor<256xf32>
+    %35 = "tt.reduce"(%34) <{axis = 0 : i32}> ({
+    ^bb0(%arg8: f32, %arg9: f32):
+      %46 = arith.addf %arg8, %arg9 : f32
+      tt.reduce.return %46 : f32
+    }) : (tensor<256xf32>) -> f32
+    %36 = arith.addf %35, %cst_0 : f32
+    %37 = arith.divf %36, %cst_1 : f32
+    %38 = arith.addf %37, %cst_2 : f32
+    %39 = tt.extern_elementwise %38 {libname = "libdevice", libpath = "/usr/local/lib/python3.10/dist-packages/triton/language/../third_party/cuda/lib/libdevice.10.bc", pure = true, symbol = "__nv_rsqrtf"} : (f32) -> f32
+    %40 = tt.splat %39 : (f32) -> tensor<256xf32>
+    %41 = arith.mulf %32, %40 : tensor<256xf32>
+    %42 = arith.mulf %41, %23 : tensor<256xf32>
+    %43 = tt.splat %arg5 : (!tt.ptr<bf16, 1>) -> tensor<256x!tt.ptr<bf16, 1>>
+    %44 = tt.addptr %43, %5 : tensor<256x!tt.ptr<bf16, 1>>, tensor<256xi32>
+    %45 = arith.truncf %42 : tensor<256xf32> to tensor<256xbf16>
+    tt.store %44, %45, %2 {cache = 1 : i32, evict = 1 : i32} : tensor<256xbf16>
+    tt.return
+  }
+}