/* Test that data values are written and read with proper alignment. */
static void
test_alignment(void)
{
void *ctx = ralloc_context(NULL);
struct blob *blob;
struct blob_reader reader;
uint8_t bytes[] = "ABCDEFGHIJKLMNOP";
size_t delta, last, num_bytes;
blob = blob_create(ctx);
/* First, write an intptr value to the blob and capture that size. This is
* the expected offset between any pair of intptr values (if written with
* alignment).
*/
blob_write_intptr(blob, (intptr_t) blob);
delta = blob->size;
last = blob->size;
/* Then loop doing the following:
*
* 1. Write an unaligned number of bytes
* 2. Verify that write results in an unaligned size
* 3. Write an intptr_t value
* 2. Verify that that write results in an aligned size
*/
for (num_bytes = 1; num_bytes < sizeof(intptr_t); num_bytes++) {
blob_write_bytes(blob, bytes, num_bytes);
expect_unequal(delta, blob->size - last, "unaligned write of bytes");
blob_write_intptr(blob, (intptr_t) blob);
expect_equal(2 * delta, blob->size - last, "aligned write of intptr");
last = blob->size;
}
/* Finally, test that reading also does proper alignment. Since we know
* that values were written with all the right alignment, all we have to do
* here is verify that correct values are read.
*/
blob_reader_init(&reader, blob->data, blob->size);
expect_equal((intptr_t) blob, blob_read_intptr(&reader),
"read of initial, aligned intptr_t");
for (num_bytes = 1; num_bytes < sizeof(intptr_t); num_bytes++) {
expect_equal_bytes(bytes, blob_read_bytes(&reader, num_bytes),
num_bytes, "unaligned read of bytes");
expect_equal((intptr_t) blob, blob_read_intptr(&reader),
"aligned read of intptr_t");
}
ralloc_free(ctx);
}
/* Test that we detect overrun. */
static void
test_overrun(void)
{
struct blob blob;
struct blob_reader reader;
uint32_t value = 0xdeadbeef;
blob_init(&blob);
blob_write_uint32(&blob, value);
blob_reader_init(&reader, blob.data, blob.size);
expect_equal(value, blob_read_uint32(&reader), "read before overrun");
expect_equal(false, reader.overrun, "overrun flag not set");
expect_equal(0, blob_read_uint32(&reader), "read at overrun");
expect_equal(true, reader.overrun, "overrun flag set");
blob_finish(&blob);
}
/* Test that we can read and write some large objects, (exercising the code in
* the blob_write functions to realloc blob->data.
*/
static void
test_big_objects(void)
{
void *ctx = ralloc_context(NULL);
struct blob blob;
struct blob_reader reader;
int size = 1000;
int count = 1000;
size_t i;
char *buf;
blob_init(&blob);
/* Initialize our buffer. */
buf = ralloc_size(ctx, size);
for (i = 0; i < size; i++) {
buf[i] = i % 256;
}
/* Write it many times. */
for (i = 0; i < count; i++) {
blob_write_bytes(&blob, buf, size);
}
blob_reader_init(&reader, blob.data, blob.size);
/* Read and verify it many times. */
for (i = 0; i < count; i++) {
expect_equal_bytes((uint8_t *) buf, blob_read_bytes(&reader, size), size,
"read of large objects");
}
expect_equal(reader.end - reader.data, reader.current - reader.data,
"number of bytes read reading large objects");
expect_equal(false, reader.overrun,
"overrun flag not set reading large objects");
blob_finish(&blob);
ralloc_free(ctx);
}
/* Test that we detect overrun. */
static void
test_overrun(void)
{
void *ctx =ralloc_context(NULL);
struct blob *blob;
struct blob_reader reader;
uint32_t value = 0xdeadbeef;
blob = blob_create(ctx);
blob_write_uint32(blob, value);
blob_reader_init(&reader, blob->data, blob->size);
expect_equal(value, blob_read_uint32(&reader), "read before overrun");
expect_equal(false, reader.overrun, "overrun flag not set");
expect_equal(0, blob_read_uint32(&reader), "read at overrun");
expect_equal(true, reader.overrun, "overrun flag set");
ralloc_free(ctx);
}
/* Test at least one call of each blob_write_foo and blob_read_foo function,
* verifying that we read out everything we wrote, that every bytes is
* consumed, and that the overrun bit is not set.
*/
static void
test_write_and_read_functions (void)
{
struct blob blob;
struct blob_reader reader;
ssize_t reserved;
size_t str_offset, uint_offset;
uint8_t reserve_buf[sizeof(reserve_test_str)];
blob_init(&blob);
/*** Test blob by writing one of every possible kind of value. */
blob_write_bytes(&blob, bytes_test_str, sizeof(bytes_test_str));
reserved = blob_reserve_bytes(&blob, sizeof(reserve_test_str));
blob_overwrite_bytes(&blob, reserved, reserve_test_str, sizeof(reserve_test_str));
/* Write a placeholder, (to be replaced later via overwrite_bytes) */
str_offset = blob.size;
blob_write_bytes(&blob, placeholder_str, sizeof(placeholder_str));
blob_write_uint32(&blob, uint32_test);
/* Write a placeholder, (to be replaced later via overwrite_uint32) */
uint_offset = blob.size;
blob_write_uint32(&blob, uint32_placeholder);
blob_write_uint64(&blob, uint64_test);
blob_write_intptr(&blob, (intptr_t) &blob);
blob_write_string(&blob, string_test_str);
/* Finally, overwrite our placeholders. */
blob_overwrite_bytes(&blob, str_offset, overwrite_test_str,
sizeof(overwrite_test_str));
blob_overwrite_uint32(&blob, uint_offset, uint32_overwrite);
/*** Now read each value and verify. */
blob_reader_init(&reader, blob.data, blob.size);
expect_equal_str(bytes_test_str,
blob_read_bytes(&reader, sizeof(bytes_test_str)),
"blob_write/read_bytes");
blob_copy_bytes(&reader, reserve_buf, sizeof(reserve_buf));
expect_equal_str(reserve_test_str, (char *) reserve_buf,
"blob_reserve_bytes/blob_copy_bytes");
expect_equal_str(overwrite_test_str,
blob_read_bytes(&reader, sizeof(overwrite_test_str)),
"blob_overwrite_bytes");
expect_equal(uint32_test, blob_read_uint32(&reader),
"blob_write/read_uint32");
expect_equal(uint32_overwrite, blob_read_uint32(&reader),
"blob_overwrite_uint32");
expect_equal(uint64_test, blob_read_uint64(&reader),
"blob_write/read_uint64");
expect_equal((intptr_t) &blob, blob_read_intptr(&reader),
"blob_write/read_intptr");
expect_equal_str(string_test_str, blob_read_string(&reader),
"blob_write/read_string");
expect_equal(reader.end - reader.data, reader.current - reader.data,
"read_consumes_all_bytes");
expect_equal(false, reader.overrun, "read_does_not_overrun");
blob_finish(&blob);
}
static void
st_deserialise_ir_program(struct gl_context *ctx,
struct gl_shader_program *shProg,
struct gl_program *prog, bool nir)
{
struct st_context *st = st_context(ctx);
size_t size = prog->driver_cache_blob_size;
uint8_t *buffer = (uint8_t *) prog->driver_cache_blob;
const struct nir_shader_compiler_options *options =
ctx->Const.ShaderCompilerOptions[prog->info.stage].NirOptions;
assert(prog->driver_cache_blob && prog->driver_cache_blob_size > 0);
struct blob_reader blob_reader;
blob_reader_init(&blob_reader, buffer, size);
switch (prog->info.stage) {
case MESA_SHADER_VERTEX: {
struct st_vertex_program *stvp = (struct st_vertex_program *) prog;
st_release_vp_variants(st, stvp);
stvp->num_inputs = blob_read_uint32(&blob_reader);
blob_copy_bytes(&blob_reader, (uint8_t *) stvp->index_to_input,
sizeof(stvp->index_to_input));
blob_copy_bytes(&blob_reader, (uint8_t *) stvp->input_to_index,
sizeof(stvp->input_to_index));
blob_copy_bytes(&blob_reader, (uint8_t *) stvp->result_to_output,
sizeof(stvp->result_to_output));
read_stream_out_from_cache(&blob_reader, &stvp->tgsi);
if (nir) {
stvp->tgsi.type = PIPE_SHADER_IR_NIR;
stvp->shader_program = shProg;
stvp->tgsi.ir.nir = nir_deserialize(NULL, options, &blob_reader);
prog->nir = stvp->tgsi.ir.nir;
} else {
read_tgsi_from_cache(&blob_reader, &stvp->tgsi.tokens,
&stvp->num_tgsi_tokens);
}
if (st->vp == stvp)
st->dirty |= ST_NEW_VERTEX_PROGRAM(st, stvp);
break;
}
case MESA_SHADER_TESS_CTRL: {
struct st_common_program *sttcp = st_common_program(prog);
st_release_basic_variants(st, sttcp->Base.Target,
&sttcp->variants, &sttcp->tgsi);
read_stream_out_from_cache(&blob_reader, &sttcp->tgsi);
if (nir) {
sttcp->tgsi.type = PIPE_SHADER_IR_NIR;
sttcp->shader_program = shProg;
sttcp->tgsi.ir.nir = nir_deserialize(NULL, options, &blob_reader);
prog->nir = sttcp->tgsi.ir.nir;
} else {
read_tgsi_from_cache(&blob_reader, &sttcp->tgsi.tokens,
&sttcp->num_tgsi_tokens);
}
if (st->tcp == sttcp)
st->dirty |= sttcp->affected_states;
break;
}
case MESA_SHADER_TESS_EVAL: {
struct st_common_program *sttep = st_common_program(prog);
st_release_basic_variants(st, sttep->Base.Target,
&sttep->variants, &sttep->tgsi);
read_stream_out_from_cache(&blob_reader, &sttep->tgsi);
if (nir) {
sttep->tgsi.type = PIPE_SHADER_IR_NIR;
sttep->shader_program = shProg;
sttep->tgsi.ir.nir = nir_deserialize(NULL, options, &blob_reader);
prog->nir = sttep->tgsi.ir.nir;
} else {
read_tgsi_from_cache(&blob_reader, &sttep->tgsi.tokens,
&sttep->num_tgsi_tokens);
}
if (st->tep == sttep)
st->dirty |= sttep->affected_states;
break;
}
case MESA_SHADER_GEOMETRY: {
struct st_common_program *stgp = st_common_program(prog);
st_release_basic_variants(st, stgp->Base.Target, &stgp->variants,
&stgp->tgsi);
read_stream_out_from_cache(&blob_reader, &stgp->tgsi);
if (nir) {
stgp->tgsi.type = PIPE_SHADER_IR_NIR;
stgp->shader_program = shProg;
stgp->tgsi.ir.nir = nir_deserialize(NULL, options, &blob_reader);
prog->nir = stgp->tgsi.ir.nir;
} else {
read_tgsi_from_cache(&blob_reader, &stgp->tgsi.tokens,
&stgp->num_tgsi_tokens);
}
if (st->gp == stgp)
st->dirty |= stgp->affected_states;
break;
}
case MESA_SHADER_FRAGMENT: {
struct st_fragment_program *stfp = (struct st_fragment_program *) prog;
st_release_fp_variants(st, stfp);
if (nir) {
stfp->tgsi.type = PIPE_SHADER_IR_NIR;
stfp->shader_program = shProg;
stfp->tgsi.ir.nir = nir_deserialize(NULL, options, &blob_reader);
prog->nir = stfp->tgsi.ir.nir;
} else {
read_tgsi_from_cache(&blob_reader, &stfp->tgsi.tokens,
&stfp->num_tgsi_tokens);
}
if (st->fp == stfp)
st->dirty |= stfp->affected_states;
break;
}
case MESA_SHADER_COMPUTE: {
struct st_compute_program *stcp = (struct st_compute_program *) prog;
st_release_cp_variants(st, stcp);
if (nir) {
stcp->tgsi.ir_type = PIPE_SHADER_IR_NIR;
stcp->shader_program = shProg;
stcp->tgsi.prog = nir_deserialize(NULL, options, &blob_reader);
prog->nir = (nir_shader *) stcp->tgsi.prog;
} else {
read_tgsi_from_cache(&blob_reader,
(const struct tgsi_token**) &stcp->tgsi.prog,
&stcp->num_tgsi_tokens);
}
stcp->tgsi.req_local_mem = stcp->Base.info.cs.shared_size;
stcp->tgsi.req_private_mem = 0;
stcp->tgsi.req_input_mem = 0;
if (st->cp == stcp)
st->dirty |= stcp->affected_states;
break;
}
default:
unreachable("Unsupported stage");
}
/* Make sure we don't try to read more data than we wrote. This should
* never happen in release builds but its useful to have this check to
* catch development bugs.
*/
if (blob_reader.current != blob_reader.end || blob_reader.overrun) {
assert(!"Invalid TGSI shader disk cache item!");
if (ctx->_Shader->Flags & GLSL_CACHE_INFO) {
fprintf(stderr, "Error reading program from cache (invalid "
"TGSI cache item)\n");
}
}
st_set_prog_affected_state_flags(prog);
_mesa_associate_uniform_storage(ctx, shProg, prog, false);
/* Create Gallium shaders now instead of on demand. */
if (ST_DEBUG & DEBUG_PRECOMPILE ||
st->shader_has_one_variant[prog->info.stage])
st_precompile_shader_variant(st, prog);
}
bool
shader_cache_read_program_metadata(struct gl_context *ctx,
struct gl_shader_program *prog)
{
/* Fixed function programs generated by Mesa are not cached. So don't
* try to read metadata for them from the cache.
*/
if (prog->Name == 0)
return false;
struct disk_cache *cache = ctx->Cache;
if (!cache)
return false;
/* Include bindings when creating sha1. These bindings change the resulting
* binary so they are just as important as the shader source.
*/
char *buf = ralloc_strdup(NULL, "vb: ");
prog->AttributeBindings->iterate(create_binding_str, &buf);
ralloc_strcat(&buf, "fb: ");
prog->FragDataBindings->iterate(create_binding_str, &buf);
ralloc_strcat(&buf, "fbi: ");
prog->FragDataIndexBindings->iterate(create_binding_str, &buf);
/* SSO has an effect on the linked program so include this when generating
* the sha also.
*/
ralloc_asprintf_append(&buf, "sso: %s\n",
prog->SeparateShader ? "T" : "F");
/* A shader might end up producing different output depending on the glsl
* version supported by the compiler. For example a different path might be
* taken by the preprocessor, so add the version to the hash input.
*/
ralloc_asprintf_append(&buf, "api: %d glsl: %d fglsl: %d\n",
ctx->API, ctx->Const.GLSLVersion,
ctx->Const.ForceGLSLVersion);
/* We run the preprocessor on shaders after hashing them, so we need to
* add any extension override vars to the hash. If we don't do this the
* preprocessor could result in different output and we could load the
* wrong shader.
*/
char *ext_override = getenv("MESA_EXTENSION_OVERRIDE");
if (ext_override) {
ralloc_asprintf_append(&buf, "ext:%s", ext_override);
}
/* DRI config options may also change the output from the compiler so
* include them as an input to sha1 creation.
*/
char sha1buf[41];
_mesa_sha1_format(sha1buf, ctx->Const.dri_config_options_sha1);
ralloc_strcat(&buf, sha1buf);
for (unsigned i = 0; i < prog->NumShaders; i++) {
struct gl_shader *sh = prog->Shaders[i];
_mesa_sha1_format(sha1buf, sh->sha1);
ralloc_asprintf_append(&buf, "%s: %s\n",
_mesa_shader_stage_to_abbrev(sh->Stage), sha1buf);
}
disk_cache_compute_key(cache, buf, strlen(buf), prog->data->sha1);
ralloc_free(buf);
size_t size;
uint8_t *buffer = (uint8_t *) disk_cache_get(cache, prog->data->sha1,
&size);
if (buffer == NULL) {
/* Cached program not found. We may have seen the individual shaders
* before and skipped compiling but they may not have been used together
* in this combination before. Fall back to linking shaders but first
* re-compile the shaders.
*
* We could probably only compile the shaders which were skipped here
* but we need to be careful because the source may also have been
* changed since the last compile so for now we just recompile
* everything.
*/
compile_shaders(ctx, prog);
return false;
}
if (ctx->_Shader->Flags & GLSL_CACHE_INFO) {
_mesa_sha1_format(sha1buf, prog->data->sha1);
fprintf(stderr, "loading shader program meta data from cache: %s\n",
sha1buf);
}
struct blob_reader metadata;
blob_reader_init(&metadata, buffer, size);
bool deserialized = deserialize_glsl_program(&metadata, ctx, prog);
if (!deserialized || metadata.current != metadata.end || metadata.overrun) {
/* Something has gone wrong discard the item from the cache and rebuild
* from source.
*/
assert(!"Invalid GLSL shader disk cache item!");
if (ctx->_Shader->Flags & GLSL_CACHE_INFO) {
fprintf(stderr, "Error reading program from cache (invalid GLSL "
"cache item)\n");
}
disk_cache_remove(cache, prog->data->sha1);
compile_shaders(ctx, prog);
free(buffer);
return false;
}
/* This is used to flag a shader retrieved from cache */
prog->data->LinkStatus = linking_skipped;
/* Since the program load was successful, CompileStatus of all shaders at
* this point should normally be compile_skipped. However because of how
* the eviction works, it may happen that some of the individual shader keys
* have been evicted, resulting in unnecessary recompiles on this load, so
* mark them again to skip such recompiles next time.
*/
char sha1_buf[41];
for (unsigned i = 0; i < prog->NumShaders; i++) {
if (prog->Shaders[i]->CompileStatus == compiled_no_opts) {
disk_cache_put_key(cache, prog->Shaders[i]->sha1);
if (ctx->_Shader->Flags & GLSL_CACHE_INFO) {
_mesa_sha1_format(sha1_buf, prog->Shaders[i]->sha1);
fprintf(stderr, "re-marking shader: %s\n", sha1_buf);
}
}
}
free (buffer);
return true;
}
