Add nhwc layout to gpu backend (#1391)

Can be enabled via environment variable MIGRAPHX_ENABLE_NHWC

src/targets/gpu/target.cpp

@@ -35,6 +35,7 @@
 #include <migraphx/fuse_pointwise.hpp>
 #include <migraphx/inline_module.hpp>
 #include <migraphx/insert_pad.hpp>
+#include <migraphx/layout_nhwc.hpp>
 #include <migraphx/memory_coloring.hpp>
 #include <migraphx/normalize_ops.hpp>
 #include <migraphx/preallocate_param.hpp>
@@ -50,6 +51,7 @@
 #include <migraphx/simplify_qdq.hpp>
 #include <migraphx/simplify_reshapes.hpp>
 #include <migraphx/gpu/allocation_model.hpp>
+#include <migraphx/gpu/compile_miopen.hpp>
 #include <migraphx/gpu/compile_ops.hpp>
 #include <migraphx/gpu/concat_gpu_opt.hpp>
 #include <migraphx/gpu/context.hpp>
@@ -70,6 +72,7 @@ namespace gpu {
 
 MIGRAPHX_DECLARE_ENV_VAR(MIGRAPHX_DISABLE_SCHEDULE_PASS)
 MIGRAPHX_DECLARE_ENV_VAR(MIGRAPHX_DISABLE_POINTWISE_FUSION)
+MIGRAPHX_DECLARE_ENV_VAR(MIGRAPHX_ENABLE_NHWC)
 
 struct id_pass
 {
@@ -120,6 +123,9 @@ std::vector<pass> target::get_passes(migraphx::context& gctx, const compile_opti
         dead_code_elimination{},
         simplify_algebra{},
         simplify_reshapes{},
+        enable_pass(enabled(MIGRAPHX_ENABLE_NHWC{}), layout_nhwc{}),
+        dead_code_elimination{},
+        simplify_reshapes{},
         simplify_algebra{},
         prefuse_ops{},
         dead_code_elimination{},
@@ -136,8 +142,12 @@ std::vector<pass> target::get_passes(migraphx::context& gctx, const compile_opti
         dead_code_elimination{},
         eliminate_concat{concat_gpu_optimization{}},
         dead_code_elimination{},
+        compile_miopen{&gctx},
+        dead_code_elimination{},
         pack_int8_args{},
         dead_code_elimination{},
+        adjust_allocation{gpu_allocation_model{}},
+        dead_code_elimination{},
         fuse_ops{&ctx, options.fast_math},
         dead_code_elimination{},
         replace_allocate{gpu_allocation_model{}, options.offload_copy},

test/layout_nhwc.cpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ */
+#include <migraphx/layout_nhwc.hpp>
+#include <migraphx/dead_code_elimination.hpp>
+#include <migraphx/pass_manager.hpp>
+#include <migraphx/operators.hpp>
+#include <migraphx/generate.hpp>
+#include <migraphx/ranges.hpp>
+#include <migraphx/instruction.hpp>
+#include <basic_ops.hpp>
+#include <migraphx/make_op.hpp>
+
+#include <test.hpp>
+
+void run_pass(migraphx::module& m)
+{
+    migraphx::run_passes(m, {migraphx::layout_nhwc{}, migraphx::dead_code_elimination{}});
+}
+
+migraphx::operation layout(std::vector<int64_t> permutation = {0, 1, 2, 3})
+{
+    return migraphx::make_op("layout", {{"permutation", permutation}});
+}
+
+migraphx::instruction_ref add_layout_nhwc(migraphx::module& m, migraphx::instruction_ref ins)
+{
+    return m.add_instruction(layout({0, 2, 3, 1}), ins);
+}
+
+TEST_CASE(conv_relu)
+{
+    migraphx::module m1;
+    {
+        auto x = m1.add_parameter("x", {migraphx::shape::float_type, {1, 8, 16, 16}});
+        auto w = m1.add_literal(
+            migraphx::generate_literal({migraphx::shape::float_type, {16, 8, 3, 3}}));
+        auto conv = m1.add_instruction(
+            migraphx::make_op("convolution",
+                              {{"padding", {1, 1}}, {"stride", {2, 2}}, {"dilation", {1, 1}}}),
+            x,
+            w);
+        m1.add_instruction(migraphx::make_op("relu"), conv);
+    }
+    run_pass(m1);
+
+    migraphx::module m2;
+    {
+        auto x = add_layout_nhwc(
+            m2, m2.add_parameter("x", {migraphx::shape::float_type, {1, 8, 16, 16}}));
+        auto w    = add_layout_nhwc(m2,
+                                 m2.add_literal(migraphx::generate_literal(
+                                     {migraphx::shape::float_type, {16, 8, 3, 3}})));
+        auto conv = m2.add_instruction(
+            migraphx::make_op("convolution",
+                              {{"padding", {1, 1}}, {"stride", {2, 2}}, {"dilation", {1, 1}}}),
+            x,
+            w);
+        auto conv_layout = m2.add_instruction(layout(), conv);
+        m2.add_instruction(migraphx::make_op("relu"), conv_layout);
+    }
+    EXPECT(m1.sort() == m2.sort());
+}
+
+TEST_CASE(conv_add)
+{
+    migraphx::module m1;
+    {
+        auto x = m1.add_parameter("x", {migraphx::shape::float_type, {1, 8, 16, 16}});
+        auto w = m1.add_literal(
+            migraphx::generate_literal({migraphx::shape::float_type, {16, 8, 3, 3}}));
+        auto y    = m1.add_literal(migraphx::generate_literal({migraphx::shape::float_type, {16}}));
+        auto conv = m1.add_instruction(
+            migraphx::make_op("convolution",
+                              {{"padding", {1, 1}}, {"stride", {2, 2}}, {"dilation", {1, 1}}}),
+            x,
+            w);
+        auto b = m1.add_instruction(
+            migraphx::make_op("broadcast", {{"axis", 1}, {"out_lens", conv->get_shape().lens()}}),
+            y);
+        m1.add_instruction(migraphx::make_op("add"), conv, b);
+    }
+    run_pass(m1);
+
+    migraphx::module m2;
+    {
+        auto x = add_layout_nhwc(
+            m2, m2.add_parameter("x", {migraphx::shape::float_type, {1, 8, 16, 16}}));
+        auto w    = add_layout_nhwc(m2,
+                                 m2.add_literal(migraphx::generate_literal(
+                                     {migraphx::shape::float_type, {16, 8, 3, 3}})));
+        auto y    = m2.add_literal(migraphx::generate_literal({migraphx::shape::float_type, {16}}));
+        auto conv = m2.add_instruction(
+            migraphx::make_op("convolution",
+                              {{"padding", {1, 1}}, {"stride", {2, 2}}, {"dilation", {1, 1}}}),
+            x,
+            w);
+        auto conv_layout = m2.add_instruction(layout(), conv);
+        auto b           = m2.add_instruction(
+            migraphx::make_op("broadcast", {{"axis", 1}, {"out_lens", conv->get_shape().lens()}}),
+            y);
+        m2.add_instruction(migraphx::make_op("add"), conv_layout, b);
+    }
+    EXPECT(m1.sort() == m2.sort());
+}
+
+int main(int argc, const char* argv[]) { test::run(argc, argv); }