writer/spirv: Fix stack-use-after-return
Never create semantic types on the stack. Always use ProgramBuilder::create(). These stack pointers have a tendency of being stored, either by other types as sub-types, or by maps (see associated bug). Also, there's a lot of logic that assumes that semantic types are de-duplicated, and that you can compare pointers. Creating new instances on the stack will break this in exciting ways. Fixed: chromium:1243944 Change-Id: I40a652f8c424030106adad2e6531287af13c8714 Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/62943 Auto-Submit: Ben Clayton <bclayton@google.com> Commit-Queue: David Neto <dneto@google.com> Kokoro: Kokoro <noreply+kokoro@google.com> Reviewed-by: David Neto <dneto@google.com>
This commit is contained in:
parent
200cdd2052
commit
e9fdd50443
|
@ -478,9 +478,9 @@ bool Builder::GenerateExecutionModes(ast::Function* func, uint32_t id) {
|
|||
}
|
||||
has_overridable_workgroup_size_ = true;
|
||||
|
||||
sem::U32 u32;
|
||||
sem::Vector vec3_u32(&u32, 3);
|
||||
uint32_t vec3_u32_type_id = GenerateTypeIfNeeded(&vec3_u32);
|
||||
auto* vec3_u32 =
|
||||
builder_.create<sem::Vector>(builder_.create<sem::U32>(), 3);
|
||||
uint32_t vec3_u32_type_id = GenerateTypeIfNeeded(vec3_u32);
|
||||
if (vec3_u32_type_id == 0) {
|
||||
return 0;
|
||||
}
|
||||
|
@ -1675,23 +1675,19 @@ uint32_t Builder::GenerateConstantIfNeeded(const ScalarConstant& constant) {
|
|||
|
||||
switch (constant.kind) {
|
||||
case ScalarConstant::Kind::kU32: {
|
||||
sem::U32 u32;
|
||||
type_id = GenerateTypeIfNeeded(&u32);
|
||||
type_id = GenerateTypeIfNeeded(builder_.create<sem::U32>());
|
||||
break;
|
||||
}
|
||||
case ScalarConstant::Kind::kI32: {
|
||||
sem::I32 i32;
|
||||
type_id = GenerateTypeIfNeeded(&i32);
|
||||
type_id = GenerateTypeIfNeeded(builder_.create<sem::I32>());
|
||||
break;
|
||||
}
|
||||
case ScalarConstant::Kind::kF32: {
|
||||
sem::F32 f32;
|
||||
type_id = GenerateTypeIfNeeded(&f32);
|
||||
type_id = GenerateTypeIfNeeded(builder_.create<sem::F32>());
|
||||
break;
|
||||
}
|
||||
case ScalarConstant::Kind::kBool: {
|
||||
sem::Bool bool_;
|
||||
type_id = GenerateTypeIfNeeded(&bool_);
|
||||
type_id = GenerateTypeIfNeeded(builder_.create<sem::Bool>());
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
@ -2457,8 +2453,9 @@ uint32_t Builder::GenerateIntrinsic(ast::CallExpression* call,
|
|||
const uint32_t kMaxNormalExponent = 0x7f00000;
|
||||
|
||||
auto set_id = GetGLSLstd450Import();
|
||||
sem::U32 u32;
|
||||
auto unsigned_id = GenerateTypeIfNeeded(&u32);
|
||||
auto* u32 = builder_.create<sem::U32>();
|
||||
|
||||
auto unsigned_id = GenerateTypeIfNeeded(u32);
|
||||
auto exponent_mask_id =
|
||||
GenerateConstantIfNeeded(ScalarConstant::U32(kExponentMask));
|
||||
auto min_exponent_id =
|
||||
|
@ -2470,8 +2467,8 @@ uint32_t Builder::GenerateIntrinsic(ast::CallExpression* call,
|
|||
// same size, and create vector constants by replicating the scalars.
|
||||
// I expect backend compilers to fold these into unique constants, so
|
||||
// there is no loss of efficiency.
|
||||
sem::Vector uvec_ty(&u32, fvec_ty->Width());
|
||||
unsigned_id = GenerateTypeIfNeeded(&uvec_ty);
|
||||
auto* uvec_ty = builder_.create<sem::Vector>(u32, fvec_ty->Width());
|
||||
unsigned_id = GenerateTypeIfNeeded(uvec_ty);
|
||||
auto splat = [&](uint32_t scalar_id) -> uint32_t {
|
||||
auto splat_result = result_op();
|
||||
OperandList splat_params{Operand::Int(unsigned_id), splat_result};
|
||||
|
@ -2562,12 +2559,12 @@ uint32_t Builder::GenerateIntrinsic(ast::CallExpression* call,
|
|||
auto* result_vector_type = intrinsic->ReturnType()->As<sem::Vector>();
|
||||
if (result_vector_type &&
|
||||
intrinsic->Parameters()[2]->Type()->is_scalar()) {
|
||||
sem::Bool bool_type;
|
||||
sem::Vector bool_vec_type(&bool_type, result_vector_type->Width());
|
||||
if (!GenerateTypeIfNeeded(&bool_vec_type)) {
|
||||
auto* bool_vec_ty = builder_.create<sem::Vector>(
|
||||
builder_.create<sem::Bool>(), result_vector_type->Width());
|
||||
if (!GenerateTypeIfNeeded(bool_vec_ty)) {
|
||||
return 0;
|
||||
}
|
||||
cond_id = GenerateSplat(cond_id, &bool_vec_type);
|
||||
cond_id = GenerateSplat(cond_id, bool_vec_ty);
|
||||
if (cond_id == 0) {
|
||||
return 0;
|
||||
}
|
||||
|
@ -3231,8 +3228,8 @@ bool Builder::GenerateAtomicIntrinsic(ast::CallExpression* call,
|
|||
return false;
|
||||
}
|
||||
|
||||
sem::Bool bool_sem_type;
|
||||
auto bool_type = GenerateTypeIfNeeded(&bool_sem_type);
|
||||
auto* bool_sem_ty = builder_.create<sem::Bool>();
|
||||
auto bool_type = GenerateTypeIfNeeded(bool_sem_ty);
|
||||
if (bool_type == 0) {
|
||||
return false;
|
||||
}
|
||||
|
@ -3924,8 +3921,7 @@ bool Builder::GenerateTextureType(const sem::Texture* texture,
|
|||
|
||||
uint32_t type_id = 0u;
|
||||
if (texture->IsAnyOf<sem::DepthTexture, sem::DepthMultisampledTexture>()) {
|
||||
sem::F32 f32;
|
||||
type_id = GenerateTypeIfNeeded(&f32);
|
||||
type_id = GenerateTypeIfNeeded(builder_.create<sem::F32>());
|
||||
} else if (auto* s = texture->As<sem::SampledTexture>()) {
|
||||
type_id = GenerateTypeIfNeeded(s->type());
|
||||
} else if (auto* ms = texture->As<sem::MultisampledTexture>()) {
|
||||
|
@ -3978,8 +3974,8 @@ bool Builder::GenerateArrayType(const sem::Array* ary, const Operand& result) {
|
|||
|
||||
bool Builder::GenerateMatrixType(const sem::Matrix* mat,
|
||||
const Operand& result) {
|
||||
sem::Vector col_type(mat->type(), mat->rows());
|
||||
auto col_type_id = GenerateTypeIfNeeded(&col_type);
|
||||
auto* col_type = builder_.create<sem::Vector>(mat->type(), mat->rows());
|
||||
auto col_type_id = GenerateTypeIfNeeded(col_type);
|
||||
if (has_error()) {
|
||||
return false;
|
||||
}
|
||||
|
|
Loading…
Reference in New Issue