Writing and running NIR unit testsΒΆ
NIR uses gtest for unit testing lowering and optimization passes. Tests should declare a class to use for all test cases:
class nir_my_pass_test : public nir_test {
protected:
nir_my_pass_test();
void run_pass(nir_reference_shader expected);
/* Resources used by the test */
}
nir_my_pass_test::nir_my_pass_test()
: nir_test::nir_test("nir_my_pass_test")
{
/* Create resources used by the test */
}
void nir_my_pass_test::run_pass(nir_reference_shader expected)
{
nir_validate_shader(b->shader, "before nir_my_pass");
NIR_PASS(_, b->shader, nir_my_pass);
check_nir_string(expected);
}
With this setup, the individual test cases can use nir_builder
to initialize a shader, run the pass that should be tested on it
and compare it against a string, containing the expected pass
output:
TEST_F(nir_my_pass_test, basic)
{
run_pass(NIR_REFERENCE_SHADER(R"(
shader: MESA_SHADER_COMPUTE
name: nir_my_pass_test
workgroup_size: 1, 1, 1
subgroup_size: 0
decl_function main () (entrypoint)
impl main {
block b0: // preds:
}
)"));
}
The expected string can be managed using the
bin/nir-test-runner.py
script which builds and runs NIR
unit tests. The output of the tests are compared against the
expectations. The runner can then optionally update the
expectations.
user@distro:~/mesa$ python bin/nir-test-runner.py -Bbuild
INFO: autodetecting backend as ninja
INFO: calculating backend command to run: /usr/bin/ninja -C /home/konstantin/dev/mesa/build
ninja: Entering directory `/home/konstantin/dev/mesa/build'
[1/1] Generating src/git_sha1.h with a custom command
diff --git a/src/compiler/nir/tests/opt_loop_tests.cpp b/src/compiler/nir/tests/opt_loop_tests.cpp
index 05d3c6357c6..0d4810c5f85 100644
--- a/src/compiler/nir/tests/opt_loop_tests.cpp
+++ b/src/compiler/nir/tests/opt_loop_tests.cpp
@@ -136,6 +136,7 @@ TEST_F(nir_opt_loop_test, opt_loop_merge_terminators_break_in_then)
check_nir_string(NIR_REFERENCE_SHADER(R"(
shader: MESA_SHADER_FRAGMENT
name: nir_opt_loop_test
+ subgroup_size: 0
decl_var shader_in INTERP_MODE_SMOOTH none int in (VARYING_SLOT_POS.x, 0, 0)
decl_var shader_out INTERP_MODE_NONE none int out (FRAG_RESULT_DEPTH.x, 0, 0)
decl_var ubo INTERP_MODE_NONE none int ubo1 (0, 0, 0)
Apply the changes listed above? [Y/n]y