arb_tessellation_shader/tes-gs-max-output: new test for TES/GS amplification

Test extremes of GS amplification under tessellation.

Acked-by: Marek Olšák <marek.olsak@amd.com>

3 files changed, 819 insertions, 0 deletions

diff --git a/tests/opengl.py b/tests/opengl.py
index 7b1747a3e..f7e408cd5 100644
--- a/tests/opengl.py
+++ b/tests/opengl.py
@@ -1531,6 +1531,7 @@ with profile.test_list.group_manager(
       override_class=BuiltInConstantsTest)
     g(['arb_tessellation_shader-large-uniforms'])
     g(['arb_tessellation_shader-layout-mismatch'])
+    g(['arb_tessellation_shader-tes-gs-max-output', '-small', '-scan', '1', '50'])
 
 # Group ARB_texture_multisample
 with profile.test_list.group_manager(
diff --git a/tests/spec/arb_tessellation_shader/CMakeLists.gl.txt b/tests/spec/arb_tessellation_shader/CMakeLists.gl.txt
index d70b00f3f..058a1bc49 100644
--- a/tests/spec/arb_tessellation_shader/CMakeLists.gl.txt
+++ b/tests/spec/arb_tessellation_shader/CMakeLists.gl.txt
@@ -18,5 +18,6 @@ piglit_add_executable (arb_tessellation_shader-invalid-primitive invalid-primiti
 piglit_add_executable (arb_tessellation_shader-minmax minmax.c)
 piglit_add_executable (arb_tessellation_shader-large-uniforms large-uniforms.c)
 piglit_add_executable (arb_tessellation_shader-layout-mismatch layout-mismatch.c)
+piglit_add_executable (arb_tessellation_shader-tes-gs-max-output tes-gs-max-output.cpp)
 
 # vim: ft=cmake:
diff --git a/tests/spec/arb_tessellation_shader/tes-gs-max-output.cpp b/tests/spec/arb_tessellation_shader/tes-gs-max-output.cpp
new file mode 100644
index 000000000..de0d2daf1
--- /dev/null
+++ b/tests/spec/arb_tessellation_shader/tes-gs-max-output.cpp
@@ -0,0 +1,817 @@
+/*
+ * Copyright (c) 2018 Advanced Micro Devices
+ *
+ * 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 (including the next
+ * paragraph) 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.
+ */
+
+/**
+ * \file tes-gs-max-output.cpp
+ *
+ * Stress the limits of what tessellation + geometry shaders can output using
+ * generic shaders with points as input and output primitives, allowing
+ * arbitrary:
+ * - number of input instances (instanced draws)
+ * - number of input patches per instance
+ * - (integer) tessellation factors
+ * - number of invocations (GS instances)
+ * - number of output vertices per invocation
+ * - number of output components per vertex
+ *
+ * Verification works by rendering points and writing to an SSBO from the
+ * fragment shader.
+ */
+
+#include "piglit-util-gl.h"
+
+#include <algorithm>
+#include <map>
+#include <set>
+#include <vector>
+
+#define WINDOW_SIZE 256
+
+PIGLIT_GL_TEST_CONFIG_BEGIN
+
+	config.supports_gl_compat_version = 32;
+	config.supports_gl_core_version = 32;
+	config.window_width = WINDOW_SIZE;
+	config.window_height = WINDOW_SIZE;
+	config.window_visual = PIGLIT_GL_VISUAL_DOUBLE | PIGLIT_GL_VISUAL_RGBA;
+	config.khr_no_error_support = PIGLIT_NO_ERRORS;
+
+PIGLIT_GL_TEST_CONFIG_END
+
+#define PASTE(x) #x
+#define STR(x) PASTE(x)
+
+struct testcase {
+	unsigned num_instances; /* draw instances */
+	unsigned num_patches; /* draw size / count */
+	unsigned tessfactor_u;
+	unsigned tessfactor_v;
+	unsigned num_invocations; /* GS invocations / instances */
+	unsigned num_outputs; /* # vertex ouput per GS invocation */
+	unsigned num_extra_components; /* # extra components per GS output vertex */
+
+	bool operator<(const testcase &o) const {
+		return std::make_tuple(num_instances, num_patches, tessfactor_u,
+				       tessfactor_v, num_invocations, num_outputs,
+				       num_extra_components)
+			<
+		       std::make_tuple(o.num_instances, o.num_patches, o.tessfactor_u,
+				       o.tessfactor_v, o.num_invocations, o.num_outputs,
+				       o.num_extra_components);
+	}
+};
+
+struct fragmentshaderkey {
+	unsigned num_extra_components;
+
+	bool operator<(const fragmentshaderkey &o) const {
+		return num_extra_components < o.num_extra_components;
+	}
+};
+
+struct geometryshaderkey {
+	unsigned num_invocations;
+	unsigned num_outputs;
+	unsigned num_extra_components;
+
+	bool operator<(const geometryshaderkey &o) const {
+		if (num_invocations < o.num_invocations)
+			return true;
+		if (num_invocations > o.num_invocations)
+			return false;
+		if (num_outputs < o.num_outputs)
+			return true;
+		if (num_outputs > o.num_outputs)
+			return false;
+		return num_extra_components < o.num_extra_components;
+	}
+};
+
+static std::map<fragmentshaderkey, GLuint> fragmentshaders;
+static std::map<geometryshaderkey, GLuint> testprograms;
+
+static const struct testcase default_testcase = {
+	.num_instances = 1,
+	.num_patches = 1,
+	.tessfactor_u = 1,
+	.tessfactor_v = 1,
+	.num_invocations = 1,
+	.num_outputs = 1,
+	.num_extra_components = 0,
+};
+
+static int32_t *buffer_copy;
+
+static const unsigned max_final_points = 2 * 1024 * 1024; /* requires 16 MiB buffer */
+static bool small = false;
+static GLuint vs_shader;
+static GLuint tcs_shader;
+static GLuint tes_shader;
+static GLuint vao;
+static GLuint ssbo;
+static std::vector<testcase> testcases;
+static std::set<testcase> testcases_set;
+static GLuint max_tessfactor;
+static GLuint max_gs_invocations;
+static GLuint max_gs_out_vertices;
+static GLuint max_gs_total_out_components;
+static GLuint max_gs_out_components;
+static unsigned max_gs_out_vertices_real;
+
+static const char vs_text[] =
+	"#version 150\n"
+	"\n"
+	"uniform int u_verts_per_instance;\n"
+	"\n"
+	"out int vs_tcs_id;\n"
+	"\n"
+	"void main() {\n"
+	"  vs_tcs_id = gl_InstanceID * u_verts_per_instance + gl_VertexID;\n"
+	"}\n";
+
+static const char tcs_text[] =
+	"#version 150\n"
+	"#extension GL_ARB_tessellation_shader : require\n"
+	"layout(vertices = 1) out;\n"
+	"\n"
+	"in int vs_tcs_id[];\n"
+	"\n"
+	"out int tcs_tes_id[];\n"
+	"\n"
+	"uniform int u_tessfactor_u;\n"
+	"uniform int u_tessfactor_v;\n"
+	"\n"
+	"void main() {\n"
+	"  tcs_tes_id[gl_InvocationID] = vs_tcs_id[gl_InvocationID];\n"
+	"  gl_TessLevelOuter[0] = u_tessfactor_v;\n"
+	"  gl_TessLevelOuter[1] = u_tessfactor_u;\n"
+	"}\n";
+
+static const char tes_text[] =
+	"#version 150\n"
+	"#extension GL_ARB_tessellation_shader : require\n"
+	"layout(isolines, equal_spacing) in;\n"
+	"\n"
+	"in int tcs_tes_id[];\n"
+	"\n"
+	"out int tes_gs_id;\n"
+	"\n"
+	"void main() {\n"
+	"  tes_gs_id = tcs_tes_id[0];\n"
+	"  gl_Position.x = gl_TessCoord[0];\n"
+	"  gl_Position.y = gl_TessCoord[1];\n"
+	"}\n";
+
+/* Those numbers really don't matter much for what we're trying to do here. */
+#define GEN_SEQUENCE \
+	"int seq_next(int x) {\n" \
+	"  x = (x + 1) * 709900053;\n" \
+	"  x = x ^ (x >> 17);\n" \
+	"  return x;\n" \
+	"}\n"
+
+static const char gs_text[] =
+	"#version 150\n"
+	"#extension GL_ARB_gpu_shader5 : require\n"
+	"\n"
+	"#define NUM_INVOCATIONS %d\n"
+	"#define NUM_OUT_VERTICES %d\n"
+	"#define NUM_EXTRA_COMPONENTS %d\n"
+	"\n"
+	"layout(lines, invocations = NUM_INVOCATIONS) in;\n"
+	"layout(points, max_vertices = NUM_OUT_VERTICES) out;\n"
+	"\n"
+	"uniform int u_tessfactor_u;\n"
+	"uniform int u_tessfactor_v;\n"
+	"\n"
+	"in int tes_gs_id[];\n"
+	"\n"
+	"flat out int gs_ps_data[1 + NUM_EXTRA_COMPONENTS];\n"
+	"\n"
+	GEN_SEQUENCE
+	"\n"
+	"void main() {\n"
+	"  int in_id = tes_gs_id[0] * u_tessfactor_u * u_tessfactor_v;\n"
+	"  float v = gl_in[0].gl_Position.y;\n"
+	"  in_id += u_tessfactor_u * int(v * u_tessfactor_v + 0.5);\n"
+	"  float u = min(gl_in[0].gl_Position.x, gl_in[1].gl_Position.x);\n"
+	"  in_id += int(u * u_tessfactor_u + 0.5);\n"
+	"\n"
+	"  for (int i = 0; i < NUM_OUT_VERTICES; ++i) {\n"
+	"    uint id = (in_id * NUM_INVOCATIONS + gl_InvocationID) * NUM_OUT_VERTICES + i;\n"
+	"    uint x = id %% " STR(WINDOW_SIZE) "u;\n"
+	"    uint y = (id / " STR(WINDOW_SIZE) "u) %% " STR(WINDOW_SIZE) "u;\n"
+	"    gl_Position.x = (float(x) + 0.5) / " STR(WINDOW_SIZE) " * 2.0 - 1.0;\n"
+	"    gl_Position.y = (float(y) + 0.5) / " STR(WINDOW_SIZE) " * 2.0 - 1.0;\n"
+	"    gl_Position.z = 0.0;\n"
+	"    gl_Position.w = 1.0;\n"
+	"\n"
+	"    int val = int(id);\n"
+	"    for (int j = 0; j <= NUM_EXTRA_COMPONENTS; ++j) {\n"
+	"      gs_ps_data[j] = val;\n"
+	"      val = seq_next(val);\n"
+	"    }\n"
+	"\n"
+	"    EmitVertex();\n"
+	"  }\n"
+	"}\n";
+
+static const char fs_text[] =
+	"#version 150\n"
+	"#extension GL_ARB_shader_storage_buffer_object : require\n"
+	"\n"
+	"#define NUM_EXTRA_COMPONENTS %d\n"
+	"\n"
+	"flat in int gs_ps_data[1 + NUM_EXTRA_COMPONENTS];\n"
+	"out vec4 out_color;\n"
+	"\n"
+	"layout(std430, binding = 0) buffer SSBO {\n"
+	"  ivec2 data[];\n"
+	"} ssbo;\n"
+	"\n"
+	GEN_SEQUENCE
+	"\n"
+	"void main() {\n"
+	"  int id = gs_ps_data[0];\n"
+	"  int screen_id = int(gl_FragCoord.y) * " STR(WINDOW_SIZE) " + int(gl_FragCoord.x);\n"
+	"  if (screen_id != id %% (" STR(WINDOW_SIZE * WINDOW_SIZE) ")) {\n"
+	"    ssbo.data[id].x = 1000;\n"
+	"    ssbo.data[id].y = screen_id;\n"
+	"    out_color = vec4(0.1, 0, 0, 1);\n"
+	"    return;\n"
+	"  }\n"
+	"\n"
+	"  int val = id;\n"
+	"  for (int j = 0; j <= NUM_EXTRA_COMPONENTS; ++j) {\n"
+	"    if (val != gs_ps_data[j]) {\n"
+	"      ssbo.data[id].x = 2000 + j;\n"
+	"      ssbo.data[id].y = gs_ps_data[j];\n"
+	"      out_color = vec4(0, 0.1, 0, 1);\n"
+	"      return;\n"
+	"    }\n"
+	"    val = seq_next(val);\n"
+	"  }\n"
+	"\n"
+	"  ssbo.data[id].x = 1;\n"
+	"  out_color = vec4(0, 0, 0, 1);\n"
+	"}\n";
+
+static void
+print_testcase(const struct testcase *tc)
+{
+	printf("Case: instances = %u patches = %u tessfactor = %u,%u invocations = %u "
+	       "outputs = %u extra_components = %u\n",
+	       tc->num_instances, tc->num_patches, tc->tessfactor_u, tc->tessfactor_v,
+	       tc->num_invocations, tc->num_outputs, tc->num_extra_components);
+}
+
+static void
+add_testcase(const struct testcase *tc)
+{
+	if (!testcases_set.insert(*tc).second)
+		return;
+
+	if (tc->num_instances > 64 * 1024 ||
+	    tc->num_patches > 64 * 1024 ||
+	    tc->tessfactor_u > 64 * 1024 ||
+	    tc->tessfactor_v > 64 * 1024 ||
+	    tc->num_invocations > 64 * 1024 ||
+	    tc->num_outputs > 64 * 1024 ||
+	    tc->num_extra_components > 64 * 1024) {
+		fprintf(stderr, "Excessive test case size. Are you sure?\n");
+		print_testcase(tc);
+		exit(1);
+	}
+
+	/* Multiple separate multiplies to avoid integer wraparound */
+	if ((uint64_t)tc->num_instances * tc->num_patches * tc->tessfactor_u > max_final_points ||
+	    (uint64_t)tc->tessfactor_v * tc->num_invocations * tc->num_outputs > max_final_points ||
+	    (uint64_t)tc->num_instances * tc->num_patches * tc->tessfactor_u *
+	    tc->tessfactor_v * tc->num_invocations * tc->num_outputs > max_final_points) {
+		fprintf(stderr, "Test case has more than %u final points.\n", max_final_points);
+		print_testcase(tc);
+		exit(1);
+	}
+
+	/* Check against implementation-defined limits. */
+	if (tc->tessfactor_u > max_tessfactor || tc->tessfactor_v > max_tessfactor) {
+		fprintf(stderr, "Tessellation factor too high (max: %u)\n",
+			max_tessfactor);
+		print_testcase(tc);
+		exit(1);
+	}
+	if (tc->num_outputs > max_gs_out_vertices) {
+		fprintf(stderr, "Too many output vertices (max: %d)\n",
+			max_gs_out_vertices);
+		print_testcase(tc);
+		exit(1);
+	}
+	if (tc->num_outputs * (5 + tc->num_extra_components) > max_gs_total_out_components) {
+		fprintf(stderr, "Too many output components (max: %d)\n",
+			max_gs_total_out_components);
+		print_testcase(tc);
+		exit(1);
+	}
+	if (tc->num_invocations > max_gs_invocations) {
+		fprintf(stderr, "Too many GS invocations (max: %d)\n",
+			max_gs_invocations);
+		print_testcase(tc);
+		exit(1);
+	}
+
+	/* Compile GS shader and link program */
+	geometryshaderkey gskey;
+	gskey.num_invocations = tc->num_invocations;
+	gskey.num_outputs = tc->num_outputs;
+	gskey.num_extra_components = tc->num_extra_components;
+	if (testprograms.find(gskey) == testprograms.end()) {
+		char *text;
+
+		fragmentshaderkey fskey;
+		fskey.num_extra_components = tc->num_extra_components;
+		auto fsit = fragmentshaders.find(fskey);
+		if (fsit == fragmentshaders.end()) {
+			if (asprintf(&text, fs_text, tc->num_extra_components) < 0)
+				abort();
+			GLuint fs_shader =
+				piglit_compile_shader_text(GL_FRAGMENT_SHADER, text);
+			free(text);
+
+			fsit = fragmentshaders.insert(std::make_pair(fskey, fs_shader)).first;
+		}
+
+		if (asprintf(&text, gs_text, tc->num_invocations, tc->num_outputs,
+			tc->num_extra_components) < 0)
+			abort();
+		GLuint gs_shader =
+			piglit_compile_shader_text(GL_GEOMETRY_SHADER, text);
+		free(text);
+
+		GLuint prog =  glCreateProgram();
+		glAttachShader(prog, vs_shader);
+		glAttachShader(prog, tcs_shader);
+		glAttachShader(prog, tes_shader);
+		glAttachShader(prog, gs_shader);
+		glAttachShader(prog, fsit->second);
+		glLinkProgram(prog);
+		if (!piglit_link_check_status(prog))
+			piglit_report_result(PIGLIT_FAIL);
+
+		glDeleteShader(gs_shader);
+
+		testprograms.insert(std::make_pair(gskey, prog));
+	}
+
+	testcases.push_back(*tc);
+}
+
+static bool
+run_testcase(const struct testcase *tc)
+{
+	unsigned final_points = tc->num_instances * tc->num_patches * tc->tessfactor_u *
+				tc->tessfactor_v * tc->num_invocations * tc->num_outputs;
+	unsigned bufsize = 2 * sizeof(int32_t) * final_points;
+
+	print_testcase(tc);
+
+	glClearColor(0, 0, 0, 1);
+	glClear(GL_COLOR_BUFFER_BIT);
+
+	geometryshaderkey gskey;
+	gskey.num_invocations = tc->num_invocations;
+	gskey.num_outputs = tc->num_outputs;
+	gskey.num_extra_components = tc->num_extra_components;
+	auto progit = testprograms.find(gskey);
+	assert(progit != testprograms.end());
+
+	glUseProgram(progit->second);
+	glPatchParameteri(GL_PATCH_VERTICES, 1);
+	glUniform1i(glGetUniformLocation(progit->second, "u_tessfactor_u"),
+		    tc->tessfactor_u);
+	glUniform1i(glGetUniformLocation(progit->second, "u_tessfactor_v"),
+		    tc->tessfactor_v);
+	glUniform1i(glGetUniformLocation(progit->second, "u_verts_per_instance"),
+		    tc->num_patches);
+
+	glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 0, ssbo);
+
+	memset(buffer_copy, 0, bufsize);
+	glBufferSubData(GL_SHADER_STORAGE_BUFFER, 0, bufsize, buffer_copy);
+
+	glEnable(GL_BLEND);
+	glBlendFunc(GL_ONE, GL_ONE);
+
+	glDrawArraysInstanced(GL_PATCHES, 0, tc->num_patches, tc->num_instances);
+
+	glDisable(GL_BLEND);
+
+	if (!piglit_check_gl_error(GL_NO_ERROR))
+		return false;
+
+	static const float expected[] = { 0, 0, 0, 1 };
+	bool ok = true;
+
+	if (!piglit_probe_rect_rgba(0, 0, WINDOW_SIZE, WINDOW_SIZE, expected))
+		ok = false;
+
+	glGetBufferSubData(GL_SHADER_STORAGE_BUFFER, 0, bufsize, buffer_copy);
+
+	for (unsigned i = 0; i < final_points; ++i) {
+		if (buffer_copy[2 * i] != 1) {
+			printf("Error @ %d: %d %d\n", i,
+			       buffer_copy[2 * i], buffer_copy[2 * i + 1]);
+			ok = false;
+		}
+	}
+
+	return ok;
+}
+
+
+static void
+generate_testcases_max2(const testcase &tc, bool explicit_instances, bool explicit_patches)
+{
+	unsigned amplify = tc.tessfactor_u * tc.tessfactor_v * tc.num_invocations * tc.num_outputs;
+	unsigned target_in = max_final_points / amplify;
+
+	if (small)
+		target_in = MIN2(4, target_in);
+
+	if (!explicit_instances) {
+		testcase tc1 = tc;
+		tc1.num_instances = MAX2(1, target_in / tc1.num_patches);
+		add_testcase(&tc1);
+	}
+
+	if (!explicit_patches) {
+		testcase tc1 = tc;
+		tc1.num_patches = MAX2(1, target_in / tc1.num_instances);
+		add_testcase(&tc1);
+	}
+
+	if (!explicit_instances && !explicit_patches) {
+		testcase tc1 = tc;
+		tc1.num_instances = MAX2(1, (unsigned)sqrt(target_in));
+		tc1.num_patches = MAX2(1, target_in / tc1.num_instances);
+		add_testcase(&tc1);
+	}
+
+	if (explicit_instances && explicit_patches)
+		add_testcase(&tc);
+}
+
+static void
+generate_testcases_max1(const testcase &tc, bool explicit_instances, bool explicit_patches,
+			bool explicit_tessfactor_u, bool explicit_tessfactor_v,
+			unsigned tess_out_max)
+{
+	if (!explicit_tessfactor_u) {
+		testcase tc1 = tc;
+		tc1.tessfactor_u = MIN2(MAX2(1, tess_out_max / tc1.tessfactor_v),
+					max_tessfactor);
+		generate_testcases_max2(tc1, explicit_instances, explicit_patches);
+	}
+
+	if (!explicit_tessfactor_v) {
+		testcase tc1 = tc;
+		tc1.tessfactor_v = MIN2(MAX2(1, tess_out_max / tc1.tessfactor_u),
+					max_tessfactor);
+		generate_testcases_max2(tc1, explicit_instances, explicit_patches);
+	}
+
+	if (!explicit_tessfactor_u && !explicit_tessfactor_v) {
+		testcase tc1 = tc;
+		tc1.tessfactor_u = MIN2(MAX2(1, (unsigned)sqrt(tess_out_max)),
+					max_tessfactor);
+		tc1.tessfactor_v = MIN2(MAX2(1, tess_out_max / tc1.tessfactor_u),
+					max_tessfactor);
+		generate_testcases_max2(tc1, explicit_instances, explicit_patches);
+	}
+
+	if (explicit_tessfactor_u && explicit_tessfactor_v)
+		generate_testcases_max2(tc, explicit_instances, explicit_patches);
+}
+
+static void
+generate_testcases_max(const testcase &tc, bool explicit_instances, bool explicit_patches,
+		       bool explicit_tessfactor_u, bool explicit_tessfactor_v)
+{
+	unsigned amplify = tc.num_invocations * tc.num_outputs;
+	unsigned tess_out_max = max_final_points / amplify;
+
+	if (small) {
+		generate_testcases_max1(tc, explicit_instances, explicit_patches,
+					explicit_tessfactor_u, explicit_tessfactor_v,
+					MIN2(4, tess_out_max));
+	} else {
+		generate_testcases_max1(tc, explicit_instances, explicit_patches,
+					explicit_tessfactor_u, explicit_tessfactor_v,
+					tess_out_max);
+		while (tess_out_max > 4) {
+			tess_out_max = sqrt(tess_out_max);
+			generate_testcases_max1(tc, explicit_instances, explicit_patches,
+						explicit_tessfactor_u, explicit_tessfactor_v,
+						tess_out_max);
+		}
+	}
+}
+
+static float
+rand_subdivide(int partitions)
+{
+	double x = 1.0;
+
+	for (int i = 1; i < partitions; ++i)
+		x = std::min(x, (double)rand() / ((double)RAND_MAX + 1));
+
+	return x;
+}
+
+void
+piglit_init(int argc, char **argv)
+{
+	bool explicit_instances = false;
+	bool explicit_patches = false;
+	bool explicit_tessfactor_u = false;
+	bool explicit_tessfactor_v = false;
+	bool explicit_invocations = false;
+	bool explicit_outputs = false;
+	bool explicit_components = false;
+	bool scan_mode = false;
+	unsigned scan_seed = 0;
+	unsigned scan_count = 0;
+	struct testcase explicit_testcase;
+
+	piglit_require_extension("GL_ARB_tessellation_shader");
+	piglit_require_extension("GL_ARB_shader_storage_buffer_object");
+
+	memcpy(&explicit_testcase, &default_testcase, sizeof(explicit_testcase));
+
+	int i;
+	for (i = 1; i < argc; ++i) {
+		if (!strcmp(argv[i], "-small")) {
+			small = true;
+		} else {
+			if (i + 1 >= argc)
+				break;
+
+			if (!strcmp(argv[i], "-instances")) {
+				explicit_testcase.num_instances = atoi(argv[i + 1]);
+				explicit_instances = true;
+				i++;
+			} else if (!strcmp(argv[i], "-patches")) {
+				explicit_testcase.num_patches = atoi(argv[i + 1]);
+				explicit_patches = true;
+				i++;
+			} else if (!strcmp(argv[i], "-tessfactor_u")) {
+				explicit_testcase.tessfactor_u = atoi(argv[i + 1]);
+				explicit_tessfactor_u = true;
+				i++;
+			} else if (!strcmp(argv[i], "-tessfactor_v")) {
+				explicit_testcase.tessfactor_v = atoi(argv[i + 1]);
+				explicit_tessfactor_v = true;
+				i++;
+			} else if (!strcmp(argv[i], "-invocations")) {
+				explicit_testcase.num_invocations = atoi(argv[i + 1]);
+				explicit_invocations = true;
+				i++;
+			} else if (!strcmp(argv[i], "-outputs")) {
+				explicit_testcase.num_outputs = atoi(argv[i + 1]);
+				explicit_outputs = true;
+				i++;
+			} else if (!strcmp(argv[i], "-components")) {
+				explicit_testcase.num_extra_components = atoi(argv[i + 1]);
+				explicit_components = true;
+				i++;
+			} else {
+				if (i + 2 >= argc)
+					break;
+
+				if (!strcmp(argv[i], "-scan")) {
+					scan_seed = atoi(argv[i + 1]);
+					scan_count = atoi(argv[i + 2]);
+					scan_mode = true;
+					i += 2;
+				} else
+					break;
+			}
+		}
+	}
+	if (i < argc) {
+		fprintf(stderr, "Unknown argument or too few params: %s\n", argv[i]);
+		exit(1);
+	}
+
+	/* Various GL objects needed by the test */
+	vs_shader = piglit_compile_shader_text(GL_VERTEX_SHADER, vs_text);
+	if (!piglit_check_gl_error(GL_NO_ERROR))
+		piglit_report_result(PIGLIT_FAIL);
+
+	tcs_shader = piglit_compile_shader_text(GL_TESS_CONTROL_SHADER, tcs_text);
+	if (!piglit_check_gl_error(GL_NO_ERROR))
+		piglit_report_result(PIGLIT_FAIL);
+
+	tes_shader = piglit_compile_shader_text(GL_TESS_EVALUATION_SHADER, tes_text);
+	if (!piglit_check_gl_error(GL_NO_ERROR))
+		piglit_report_result(PIGLIT_FAIL);
+
+	glGenVertexArrays(1, &vao);
+	glBindVertexArray(vao);
+
+	glGenBuffers(1, &ssbo);
+	glBindBuffer(GL_SHADER_STORAGE_BUFFER, ssbo);
+	glBufferData(GL_SHADER_STORAGE_BUFFER, max_final_points * 8, NULL, GL_DYNAMIC_READ);
+
+	buffer_copy = (int32_t *)calloc(2 * sizeof(int32_t), max_final_points);
+
+	glGetIntegerv(GL_MAX_TESS_GEN_LEVEL, (GLint*)&max_tessfactor);
+	glGetIntegerv(GL_MAX_GEOMETRY_OUTPUT_VERTICES, (GLint*)&max_gs_out_vertices);
+	glGetIntegerv(GL_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS,
+		      (GLint*)&max_gs_total_out_components);
+	glGetIntegerv(GL_MAX_GEOMETRY_OUTPUT_COMPONENTS,
+		      (GLint*)&max_gs_out_components);
+	glGetIntegerv(GL_MAX_GEOMETRY_SHADER_INVOCATIONS, (GLint*)&max_gs_invocations);
+	if (!piglit_check_gl_error(GL_NO_ERROR))
+		piglit_report_result(PIGLIT_FAIL);
+
+	max_gs_out_vertices_real = MIN2(max_gs_out_vertices,
+					max_gs_total_out_components / 5);
+
+	if (scan_mode) {
+		srand(scan_seed);
+
+		/* First, generate test cases that max out each of the dimensions */
+		testcase tc0 = explicit_testcase;
+		if (!explicit_invocations)
+			tc0.num_invocations = max_gs_invocations;
+
+		if (!explicit_outputs) {
+			testcase tc1 = tc0;
+
+			if (!explicit_components) {
+				tc1.num_outputs = max_gs_out_vertices_real;
+				tc1.num_extra_components =
+					MIN2(max_gs_total_out_components / tc1.num_outputs,
+					     max_gs_out_components) - 5;
+			} else {
+				tc1.num_outputs =
+					MIN2(max_gs_total_out_components /
+					     (5 + tc1.num_extra_components),
+					     max_gs_out_vertices_real);
+			}
+
+			generate_testcases_max(tc1, explicit_instances, explicit_patches,
+					       explicit_tessfactor_u, explicit_tessfactor_v);
+		}
+
+		if (!explicit_components) {
+			testcase tc1 = tc0;
+
+			if (!explicit_outputs) {
+				tc1.num_extra_components = max_gs_out_components - 5;
+				tc1.num_outputs =
+					MIN2(max_gs_total_out_components /
+					     (5 + tc1.num_extra_components),
+					     max_gs_out_vertices_real);
+			} else {
+				tc1.num_extra_components =
+					MIN2(max_gs_total_out_components / tc1.num_outputs,
+					     max_gs_out_components) - 4;
+			}
+
+			generate_testcases_max(tc1, explicit_instances, explicit_patches,
+					       explicit_tessfactor_u, explicit_tessfactor_v);
+		}
+
+		if (explicit_outputs && explicit_components)
+			generate_testcases_max(tc0, explicit_instances, explicit_patches,
+					       explicit_tessfactor_u, explicit_tessfactor_v);
+
+		/* Generate additional tests randomly.
+		 *
+		 * Attempt to generate a random distribution that isn't too
+		 * lop-sided, but admittedly this is all just hand-wavey
+		 * heuristics.
+		 */
+		while (testcases.size() < scan_count) {
+			testcase tc = explicit_testcase;
+
+			if (!explicit_outputs || !explicit_components) {
+				if (explicit_outputs || rand() & 1) {
+					unsigned max_components =
+						MIN2(max_gs_total_out_components / tc.num_outputs,
+						     max_gs_out_components) - 5;
+					tc.num_extra_components = rand() % (max_components + 1);
+
+					if (!explicit_outputs) {
+						unsigned max_outputs =
+							MIN2(max_gs_total_out_components /
+							     (5 + tc.num_extra_components),
+							     max_gs_out_vertices_real);
+						tc.num_outputs = 1 + rand() % max_outputs;
+					}
+				} else {
+					unsigned max_outputs =
+						MIN2(max_gs_total_out_components /
+						     (5 + tc.num_extra_components),
+						     max_gs_out_vertices_real);
+					tc.num_outputs = 1 + rand() % max_outputs;
+
+					if (!explicit_components) {
+						unsigned max_components =
+							MIN2(max_gs_total_out_components / tc.num_outputs,
+							     max_gs_out_components) - 5;
+						tc.num_extra_components = rand() % (max_components + 1);
+					}
+				}
+			}
+
+			unsigned amplify = tc.num_outputs;
+
+			if (explicit_invocations)
+				amplify *= tc.num_invocations;
+			if (explicit_tessfactor_u)
+				amplify *= tc.tessfactor_u;
+			if (explicit_tessfactor_v)
+				amplify *= tc.tessfactor_v;
+			if (explicit_patches)
+				amplify *= tc.num_patches;
+			if (explicit_instances)
+				amplify *= tc.num_instances;
+
+			unsigned target = max_final_points / amplify;
+
+			if (small)
+				target = MIN2(32, target);
+
+			if (!explicit_tessfactor_u) {
+				float tf_log_weight = logf(target) * rand_subdivide(6);
+				tc.tessfactor_u = MIN2(expf(tf_log_weight), max_tessfactor);
+				target /= tc.tessfactor_u;
+			}
+			if (!explicit_tessfactor_v) {
+				float tf_log_weight = logf(target) * rand_subdivide(6);
+				tc.tessfactor_v = MIN2(expf(tf_log_weight), max_tessfactor);
+				target /= tc.tessfactor_v;
+			}
+			if (!explicit_invocations) {
+				float log_weight = logf(target);
+				if (!explicit_instances || !explicit_patches)
+					log_weight *= rand_subdivide(2);
+				tc.num_invocations = MIN2(expf(log_weight), max_gs_invocations);
+				target /= tc.num_invocations;
+			}
+			if (!explicit_instances) {
+				float log_weight = logf(target);
+				if (!explicit_patches)
+					log_weight *= rand_subdivide(2);
+				tc.num_instances = expf(log_weight);
+				target /= tc.num_instances;
+			}
+			if (!explicit_patches)
+				tc.num_patches = 1 + rand() % target;
+
+			add_testcase(&tc);
+		}
+	} else {
+		add_testcase(&explicit_testcase);
+	}
+}
+
+enum piglit_result
+piglit_display(void)
+{
+	bool pass = true;
+
+	for (unsigned i = 0; i < testcases.size(); ++i) {
+		if (!run_testcase(&testcases[i]))
+			pass = false;
+	}
+
+	if (!piglit_check_gl_error(GL_NO_ERROR))
+		pass = false;
+
+	piglit_present_results();
+
+	return pass ? PIGLIT_PASS : PIGLIT_FAIL;
+}