Add RECIPROCAL pointwise op support

include/fusilli/attributes/pointwise_attributes.h

@@ -56,7 +56,7 @@ namespace fusilli {
   /* OP(MIN_OP) */                                                             \
   OP(MUL)                                                                      \
   /* OP(NEG) */                                                                \
-  /* OP(RECIPROCAL) */                                                         \
+  OP(RECIPROCAL)                                                               \
   /* OP(RELU_BWD) */                                                           \
   OP(RELU_FWD)                                                                 \
   /* OP(RSQRT) */                                                              \
@@ -139,6 +139,7 @@ inline const std::unordered_map<PointwiseAttr::Mode, int>
         {PointwiseAttr::Mode::CMP_NEQ, 2},
         {PointwiseAttr::Mode::DIV, 2},
         {PointwiseAttr::Mode::MUL, 2},
+        {PointwiseAttr::Mode::RECIPROCAL, 1},
         {PointwiseAttr::Mode::RELU_FWD, 1},
         {PointwiseAttr::Mode::SIGMOID_FWD, 1},
         {PointwiseAttr::Mode::SUB, 2},

include/fusilli/support/asm_emitter.h

@@ -1741,6 +1741,7 @@ inline ErrorOr<std::string> PointwiseNode::emitNodePreAsm() const {
     FUSILLI_DECLARE_BINARY_TORCH_EMITTER(CMP_NEQ, torch.aten.ne.Tensor)
     FUSILLI_DECLARE_BINARY_TORCH_EMITTER(DIV, torch.aten.div.Tensor)
     FUSILLI_DECLARE_BINARY_TORCH_EMITTER(MUL, torch.aten.mul.Tensor)
+    FUSILLI_DECLARE_UNARY_TORCH_EMITTER(RECIPROCAL, torch.aten.reciprocal)
     FUSILLI_DECLARE_UNARY_TORCH_EMITTER(RELU_FWD, torch.aten.relu)
     FUSILLI_DECLARE_UNARY_TORCH_EMITTER(SIGMOID_FWD, torch.aten.sigmoid)
     FUSILLI_DECLARE_UNARY_TORCH_EMITTER(TANH_FWD, torch.aten.tanh)

tests/CMakeLists.txt

@@ -155,6 +155,7 @@ add_fusilli_lit_tests(
     lit/test_pointwise_asm_emitter_div.cpp
     lit/test_pointwise_asm_emitter_mul.cpp
     lit/test_pointwise_asm_emitter_mul_scalar.cpp
+    lit/test_pointwise_asm_emitter_reciprocal.cpp
     lit/test_pointwise_asm_emitter_sigmoid.cpp
     lit/test_pointwise_asm_emitter_tanh.cpp
     lit/test_pointwise_asm_emitter_sub.cpp

tests/lit/test_pointwise_asm_emitter_reciprocal.cpp

@@ -0,0 +1,60 @@
+// Copyright 2026 Advanced Micro Devices, Inc.
+//
+// Licensed under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+// RUN: %{TEST_EXE} | iree-opt --verify-roundtrip
+// RUN: %{TEST_EXE} | FileCheck %s --check-prefix=TORCH-CHECK
+// RUN: %{TEST_EXE} stats | FileCheck %s --check-prefix=%{BACKEND}-STATS-CHECK
+
+// clang-format off
+//
+// TORCH-CHECK:   module @module {
+// TORCH-CHECK:     func.func @main(%[[RESULT0:.+]]: !torch.tensor<[16,256,64,32],f32>, %[[ARG0:.+]]: !torch.vtensor<[16,256,64,32],f32>) attributes {torch.assume_strict_symbolic_shapes} {
+// TORCH-CHECK:       %[[PERM0_0:.+]] = torch.constant.int 0
+// TORCH-CHECK:       %[[PERM0_1:.+]] = torch.constant.int 1
+// TORCH-CHECK:       %[[PERM0_2:.+]] = torch.constant.int 2
+// TORCH-CHECK:       %[[PERM0_3:.+]] = torch.constant.int 3
+// TORCH-CHECK:       %[[PERM0_LIST:.+]] = torch.prim.ListConstruct %[[PERM0_0]], %[[PERM0_1]], %[[PERM0_2]], %[[PERM0_3]] : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
+// TORCH-CHECK:       %[[PERMUTE0:.+]] = torch.aten.permute %[[ARG0]], %[[PERM0_LIST]] : !torch.vtensor<[16,256,64,32],f32>, !torch.list<int> -> !torch.vtensor<[16,256,64,32],f32>
+// TORCH-CHECK:       %[[RECIPROCAL:.+]] = torch.aten.reciprocal %[[PERMUTE0]] : !torch.vtensor<[16,256,64,32],f32> -> !torch.vtensor<[16,256,64,32],f32>
+// TORCH-CHECK:       %[[PERM_OUT_0:.+]] = torch.constant.int 0
+// TORCH-CHECK:       %[[PERM_OUT_1:.+]] = torch.constant.int 1
+// TORCH-CHECK:       %[[PERM_OUT_2:.+]] = torch.constant.int 2
+// TORCH-CHECK:       %[[PERM_OUT_3:.+]] = torch.constant.int 3
+// TORCH-CHECK:       %[[PERM_OUT_LIST:.+]] = torch.prim.ListConstruct %[[PERM_OUT_0]], %[[PERM_OUT_1]], %[[PERM_OUT_2]], %[[PERM_OUT_3]] : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
+// TORCH-CHECK:       %[[PERM_OUT:.+]] = torch.aten.permute %[[RECIPROCAL]], %[[PERM_OUT_LIST]] : !torch.vtensor<[16,256,64,32],f32>, !torch.list<int> -> !torch.vtensor<[16,256,64,32],f32>
+// TORCH-CHECK:       torch.overwrite.tensor.contents %[[PERM_OUT]] overwrites %[[RESULT0]] : !torch.vtensor<[16,256,64,32],f32>, !torch.tensor<[16,256,64,32],f32>
+// TORCH-CHECK:       return
+// TORCH-CHECK:     }
+// TORCH-CHECK:   }
+//
+// AMDGPU-STATS-CHECK: "transient-memory-size": 0
+// AMDGPU-STATS-CHECK: "dispatch-count": 1
+// CPU-STATS-CHECK: "transient-memory-size": 0
+// CPU-STATS-CHECK: "dispatch-count": 1
+//
+// clang-format on
+
+#include <fusilli.h>
+
+#include "utils.h"
+
+#include <iostream>
+#include <string>
+
+using namespace fusilli;
+
+int main(int argc, char **argv) {
+  std::string mode = (argc > 1) ? argv[1] : "default";
+
+  auto status = testUnaryPointwiseAsmEmitter(
+      "pointwise_asm_emitter_reciprocal", "reciprocal", mode,
+      PointwiseAttr::Mode::RECIPROCAL, {16, 256, 64, 32});
+  if (isError(status)) {
+    std::cerr << "Test failed: " << status << std::endl;
+    return 1;
+  }
+  return 0;
+}