1
0
Fork 0
forked from suyu/suyu

spirv_atomic: Define U32x2 storage buffers for 64-bit storage atomics

Some drivers do not support 64-bit atomics, and fallback to atomically modifying U32x2 vectors. This change ensures that U32x2 storage vectors are defined in the spir-v shader when 64-bit atomics are used.

Fixes a hang on some devices, notably Intel GPUs, when booting Pokemon Legends Arceus
This commit is contained in:
ameerj 2022-01-28 19:00:04 -05:00
parent 1900abde13
commit 4790ba7839
2 changed files with 3 additions and 3 deletions

View file

@ -74,7 +74,7 @@ Id StorageAtomicU64(EmitContext& ctx, const IR::Value& binding, const IR::Value&
const auto [scope, semantics]{AtomicArgs(ctx)};
return (ctx.*atomic_func)(ctx.U64, pointer, scope, semantics, value);
}
LOG_ERROR(Shader_SPIRV, "Int64 atomics not supported, fallback to non-atomic");
LOG_WARNING(Shader_SPIRV, "Int64 atomics not supported, fallback to non-atomic");
const Id pointer{StoragePointer(ctx, ctx.storage_types.U32x2, &StorageDefinitions::U32x2,
binding, offset, sizeof(u32[2]))};
const Id original_value{ctx.OpBitcast(ctx.U64, ctx.OpLoad(ctx.U32[2], pointer))};
@ -267,7 +267,7 @@ Id EmitStorageAtomicExchange64(EmitContext& ctx, const IR::Value& binding, const
const auto [scope, semantics]{AtomicArgs(ctx)};
return ctx.OpAtomicExchange(ctx.U64, pointer, scope, semantics, value);
}
LOG_ERROR(Shader_SPIRV, "Int64 atomics not supported, fallback to non-atomic");
LOG_WARNING(Shader_SPIRV, "Int64 atomics not supported, fallback to non-atomic");
const Id pointer{StoragePointer(ctx, ctx.storage_types.U32x2, &StorageDefinitions::U32x2,
binding, offset, sizeof(u32[2]))};
const Id original{ctx.OpBitcast(ctx.U64, ctx.OpLoad(ctx.U32[2], pointer))};

View file

@ -688,7 +688,7 @@ void VisitUsages(Info& info, IR::Inst& inst) {
case IR::Opcode::StorageAtomicAnd64:
case IR::Opcode::StorageAtomicOr64:
case IR::Opcode::StorageAtomicXor64:
info.used_storage_buffer_types |= IR::Type::U64;
info.used_storage_buffer_types |= IR::Type::U64 | IR::Type::U32x2;
info.uses_int64_bit_atomics = true;
break;
case IR::Opcode::BindlessImageAtomicIAdd32: