observe the results of multiply-add across implementations

- add specific test showing diverging results when using FMA
- re-write eval test to not fail due to FMA
- re-write hitcollector test to not fail due to FMA

2 files changed, 53 insertions, 0 deletions

diff --git a/eval/src/tests/eval/multiply_add/CMakeLists.txt b/eval/src/tests/eval/multiply_add/CMakeLists.txt
new file mode 100644
index 00000000000..c50aa4f50a2
--- /dev/null
+++ b/eval/src/tests/eval/multiply_add/CMakeLists.txt
@@ -0,0 +1,9 @@
+# Copyright Verizon Media. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
+vespa_add_executable(eval_multiply_add_test_app TEST
+    SOURCES
+    multiply_add_test.cpp
+    DEPENDS
+    vespaeval
+    gtest
+)
+vespa_add_test(NAME eval_multiply_add_test_app COMMAND eval_multiply_add_test_app)
diff --git a/eval/src/tests/eval/multiply_add/multiply_add_test.cpp b/eval/src/tests/eval/multiply_add/multiply_add_test.cpp
new file mode 100644
index 00000000000..35cab0a6030
--- /dev/null
+++ b/eval/src/tests/eval/multiply_add/multiply_add_test.cpp
@@ -0,0 +1,44 @@
+// Copyright Verizon Media. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
+
+#include <vespa/eval/eval/function.h>
+#include <vespa/eval/eval/llvm/compiled_function.h>
+#include <vespa/eval/eval/interpreted_function.h>
+#include <vespa/eval/tensor/default_tensor_engine.h>
+#include <vespa/vespalib/gtest/gtest.h>
+
+using namespace vespalib::eval;
+
+using Engine = vespalib::tensor::DefaultTensorEngine;
+
+double gcc_fun(double a, double b) {
+    return (a * 3) + b;
+}
+
+TEST(MultiplyAddTest, multiply_add_gives_same_result) {
+    auto fun = Function::parse("a*3+b");
+    CompiledFunction cfun(*fun, PassParams::ARRAY);
+    NodeTypes node_types = NodeTypes(*fun, {ValueType::double_type(), ValueType::double_type()});
+    InterpretedFunction ifun(Engine::ref(), *fun, node_types);
+    auto llvm_fun = cfun.get_function();
+    //-------------------------------------------------------------------------
+    double a = -1.0/3.0;
+    double b = 1.0;
+    std::vector<double> ab({a, b});
+    SimpleParams params(ab);
+    InterpretedFunction::Context ictx(ifun);
+    //-------------------------------------------------------------------------
+    const Value &result_value = ifun.eval(ictx, params);
+    double ifun_res = result_value.as_double();
+    double llvm_res = llvm_fun(&ab[0]);
+    double gcc_res = gcc_fun(a, b);
+    fprintf(stderr, "ifun_res: %a\n", ifun_res);
+    fprintf(stderr, "llvm_res: %a\n", llvm_res);
+    fprintf(stderr, "gcc_res:  %a\n", gcc_res);
+    EXPECT_EQ(ifun_res, llvm_res);
+    EXPECT_DOUBLE_EQ(llvm_res + 1.0, gcc_res + 1.0);
+    if (llvm_res != gcc_res) {
+        fprintf(stderr, "WARNING: diverging results caused by fused multiply add\n");
+    }
+}
+
+GTEST_MAIN_RUN_ALL_TESTS()