match gemm_softmax_gemm when there is no scale (#2748)

ROCm · Feb 16, 2024 · c8b6c6d · c8b6c6d
1 parent 55f79bb
commit c8b6c6d
Show file tree

Hide file tree

Showing 3 changed files with 177 additions and 16 deletions.
diff --git a/src/targets/gpu/include/migraphx/gpu/prefuse_ops.hpp b/src/targets/gpu/include/migraphx/gpu/prefuse_ops.hpp
@@ -1,7 +1,7 @@
 /*
  * The MIT License (MIT)
  *
- * Copyright (c) 2015-2023 Advanced Micro Devices, Inc. All rights reserved.
+ * Copyright (c) 2015-2024 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
@@ -36,6 +36,7 @@ namespace gpu {
 
 struct MIGRAPHX_GPU_EXPORT prefuse_ops
 {
+    bool enable_attention = false;
     std::string name() const { return "gpu::prefuse_ops"; }
     void apply(module_pass_manager& mpm) const;
 };

diff --git a/src/targets/gpu/prefuse_ops.cpp b/src/targets/gpu/prefuse_ops.cpp
@@ -1,7 +1,7 @@
 /*
  * The MIT License (MIT)
  *
- * Copyright (c) 2015-2023 Advanced Micro Devices, Inc. All rights reserved.
+ * Copyright (c) 2015-2024 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
@@ -162,36 +162,53 @@ auto is_mlir_gemm()
     });
 }
 
+auto is_test_gemm(bool enable_attention)
+{
+    return match::make_basic_pred_matcher([=](instruction_ref ins) {
+        if(ins->name() != "dot")
+            return false;
+        return enable_attention;
+    });
+}
+
 struct find_gemm_softmax_gemm
 {
+    bool enable_attention = false;
+
     auto matcher() const
     {
-        auto gemm1 = match::skip(match::name("contiguous"))(
-            match::name("dot")(match::any_of(is_ck_gemm(), is_mlir_gemm()).bind("gemm1")));
-        auto mul = match::name("mul")(
+        auto gemm1 = match::skip(match::name("contiguous"))(match::name("dot")(
+            match::any_of(is_ck_gemm(), is_mlir_gemm(), is_test_gemm(enable_attention))
+                .bind("gemm1")));
+        auto mul   = match::name("mul")(
             match::nargs(2), match::either_arg(0, 1)(match::is_constant().bind("scale"), gemm1));
-        auto softmax = match::name("softmax")(match::arg(0)(mul)).bind("softmax");
+        auto softmax =
+            match::name("softmax")(match::arg(0)(match::any_of(mul, gemm1))).bind("softmax");
 
-        return match::name("dot")(match::any_of(is_ck_gemm(), is_mlir_gemm()).bind("gemm2"))(
-            match::arg(0)(softmax));
+        return match::name("dot")(
+            match::any_of(is_ck_gemm(), is_mlir_gemm(), is_test_gemm(enable_attention))
+                .bind("gemm2"))(match::arg(0)(softmax));
     }
 
     void apply(module_pass_manager& mpm, const match::matcher_result& r) const
     {
         auto ins       = r.result;
         auto gemm2_ins = r.instructions["gemm2"];
         auto gemm1_ins = r.instructions["gemm1"];
-        auto scale_lit = r.instructions["scale"];
 
         float scale = 1.0;
-        scale_lit->eval().visit([&](const auto s) {
+        if(contains(r.instructions, "scale"))
+        {
+            auto scale_lit = r.instructions["scale"];
             // CK only supports single-valued scale
-            if(std::all_of(
-                   s.begin() + 1, s.end(), [&](auto v) { return float_equal(v, s.front()); }))
+            scale_lit->eval().visit([&](const auto s) {
+                // CK only supports single-valued scale
+                if(not std::all_of(
+                       s.begin() + 1, s.end(), [&](auto v) { return float_equal(v, s.front()); }))
+                    return;
                 scale = s.front();
-            else
-                return;
-        });
+            });
+        }
 
         auto inputs = gemm1_ins->inputs();            // A, B
         inputs.push_back(gemm2_ins->inputs().back()); // B1
@@ -208,7 +225,7 @@ void prefuse_ops::apply(module_pass_manager& mpm) const
     match::find_matches(mpm.get_module(), find_layernorm{});
     mpm.run_pass(dead_code_elimination{});
     match::find_matches(mpm.get_module(), find_add_layernorm{});
-    match::find_matches(mpm, find_gemm_softmax_gemm{});
+    match::find_matches(mpm, find_gemm_softmax_gemm{enable_attention});
 }
 
 } // namespace gpu

diff --git a/test/gpu/prefuse_ops.cpp b/test/gpu/prefuse_ops.cpp
@@ -0,0 +1,143 @@
+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2024 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/dead_code_elimination.hpp>
+#include <migraphx/gpu/prefuse_ops.hpp>
+#include <migraphx/gpu/fuse_mlir.hpp>
+#include <migraphx/gpu/gemm_softmax_gemm.hpp>
+#include <migraphx/pass_manager.hpp>
+#include <migraphx/instruction.hpp>
+#include <migraphx/program.hpp>
+#include <migraphx/generate.hpp>
+#include <basic_ops.hpp>
+#include <migraphx/make_op.hpp>
+#include <test.hpp>
+
+struct pre_gemm_softmax_gemm : migraphx::gpu::gemm_softmax_gemm
+{
+    std::string name() const { return "gpu::pre_gemm_softmax_gemm"; }
+};
+
+void run_pass(migraphx::module& m)
+{
+    migraphx::run_passes(m, {migraphx::gpu::prefuse_ops{true}, migraphx::dead_code_elimination{}});
+}
+
+TEST_CASE(find_gemm_softmax_gemm)
+{
+    migraphx::shape s1{migraphx::shape::float_type, {8, 16, 32}};
+    migraphx::shape s2{migraphx::shape::float_type, {8, 32, 16}};
+
+    migraphx::module m1;
+    {
+        auto x     = m1.add_parameter("x", s1);
+        auto y     = m1.add_parameter("y", s2);
+        auto z     = m1.add_parameter("z", s1);
+        auto scale = m1.add_literal(2.0f);
+
+        auto dot1     = m1.add_instruction(migraphx::make_op("dot"), x, y);
+        auto scale_mb = m1.add_instruction(
+            migraphx::make_op("multibroadcast", {{"out_lens", dot1->get_shape().lens()}}), scale);
+        auto mul = m1.add_instruction(migraphx::make_op("mul"), dot1, scale_mb);
+        auto sm  = m1.add_instruction(migraphx::make_op("softmax", {{"axis", 2}}), mul);
+        m1.add_instruction(migraphx::make_op("dot"), sm, z);
+    }
+    run_pass(m1);
+
+    migraphx::module m2;
+    {
+        auto x = m2.add_parameter("x", s1);
+        auto y = m2.add_parameter("y", s2);
+        auto z = m2.add_parameter("z", s1);
+
+        m2.add_instruction(pre_gemm_softmax_gemm{migraphx::make_op("dot"), 2}, x, y, z);
+    }
+
+    EXPECT(m1 == m2);
+}
+
+TEST_CASE(find_gemm_softmax_gemm_multi_scale)
+{
+    migraphx::shape s1{migraphx::shape::float_type, {8, 16, 32}};
+    migraphx::shape s2{migraphx::shape::float_type, {8, 32, 16}};
+
+    migraphx::module m1;
+    {
+        auto x = m1.add_parameter("x", s1);
+        auto y = m1.add_parameter("y", s2);
+        auto z = m1.add_parameter("z", s1);
+        auto scale =
+            m1.add_literal(migraphx::generate_literal({migraphx::shape::float_type, {16}}, 10));
+
+        auto dot1     = m1.add_instruction(migraphx::make_op("dot"), x, y);
+        auto scale_mb = m1.add_instruction(
+            migraphx::make_op("multibroadcast", {{"out_lens", dot1->get_shape().lens()}}), scale);
+        auto mul = m1.add_instruction(migraphx::make_op("mul"), dot1, scale_mb);
+        auto sm  = m1.add_instruction(migraphx::make_op("softmax", {{"axis", 2}}), mul);
+        m1.add_instruction(migraphx::make_op("dot"), sm, z);
+    }
+    run_pass(m1);
+
+    migraphx::module m2;
+    {
+        auto x = m2.add_parameter("x", s1);
+        auto y = m2.add_parameter("y", s2);
+        auto z = m2.add_parameter("z", s1);
+
+        m2.add_instruction(pre_gemm_softmax_gemm{migraphx::make_op("dot"), 1}, x, y, z);
+    }
+
+    EXPECT(m1 == m2);
+}
+
+TEST_CASE(find_gemm_softmax_gemm_no_scale)
+{
+    migraphx::shape s1{migraphx::shape::float_type, {8, 16, 32}};
+    migraphx::shape s2{migraphx::shape::float_type, {8, 32, 16}};
+
+    migraphx::module m1;
+    {
+        auto x = m1.add_parameter("x", s1);
+        auto y = m1.add_parameter("y", s2);
+        auto z = m1.add_parameter("z", s1);
+
+        auto dot1 = m1.add_instruction(migraphx::make_op("dot"), x, y);
+        auto sm   = m1.add_instruction(migraphx::make_op("softmax", {{"axis", 2}}), dot1);
+        m1.add_instruction(migraphx::make_op("dot"), sm, z);
+    }
+    run_pass(m1);
+
+    migraphx::module m2;
+    {
+        auto x = m2.add_parameter("x", s1);
+        auto y = m2.add_parameter("y", s2);
+        auto z = m2.add_parameter("z", s1);
+
+        m2.add_instruction(pre_gemm_softmax_gemm{migraphx::make_op("dot"), 1}, x, y, z);
+    }
+
+    EXPECT(m1 == m2);
+}
+
+int main(int argc, const char* argv[]) { test::run(argc, argv); }