Tags: facebook/hermes
Tags
Deploy 0.262.0 to xplat Summary: X-link: facebook/react-native#49708 Changelog: [Internal] Reviewed By: SamChou19815 Differential Revision: D70277654 fbshipit-source-id: e215e392aa6e06181f2df03ce8b9b873807f3b86
Implement more missing methods on WithRuntimeDecorator Summary: X-link: facebook/react-native#45049 WithRuntimeDecorator is missing many methods that were added after it. Add implementations for them. The underlying issue here is that because this inherits from RuntimeDecorator, which implements all methods, there is no compilation error for this runtime when we add new methods. Changelog: [GENERAL] [FIXED] - Add missing methods to the WithRuntimeDecorator class. Reviewed By: avp Differential Revision: D58752127 fbshipit-source-id: d80b4ed1c38698ed3850d0cd961bf7ddde2449a0
Add Hermes implementation for Object.create with prototype Summary: Adds the implementation `createObjectWithPrototype` JSI method for Hermes runtime. Grafted from 1053185886992b918021c4d2578a0807f797dd00 (D66487948) - Grafted path xplat/static_h to xplat/hermes Reviewed By: tmikov Differential Revision: D67412929 fbshipit-source-id: a283ffc775494f7df2bdcf9a91ada188b673c071
Deploy 0.254.2 to xplat Summary: X-link: facebook/react-native#47912 Changelog: [Internal] Reviewed By: SamChou19815 Differential Revision: D66379853 fbshipit-source-id: 3a7677c5a71ef74e509ed43be87e455aed0cbaa7
Recycle block local registers in fast pass (#1448) Summary: Original Author: neildhar@meta.com Original Git: 6b69a06 Original Reviewed By: avp Original Revision: D59072005 The register allocator has the ability to honour a memory limit that is proportional to the product of the number of instructions and basic blocks in the function being allocated. Unfortunately, functions that hit this limit by definition have a lot of instructions Even in the most degenerate case where every block has one instruction, you need 4000 instructions to hit the 10M limit. This diff tries to improve the quality of generated code in cases where most values are used within the basic block they are defined in. In such cases, we currently make the register available after the end of the block. With this diff, the registers become available after their last use in the block. This is useful for functions with extremely large basic blocks, where the current approach would end up allocating a huge number of registers since the registers cannot be used within the same block. Closes #1448 Reviewed By: avp Differential Revision: D60241766 fbshipit-source-id: 5196333862517cd546d675cf8fe005eb1ed5a790
JSError: look for stack data in the prototype chain Summary: Imported from static_h Original Author: tmikov@meta.com Original Git: 8b7a9f8 Original Reviewed By: avp Original Revision: D61870728 Look for stack data in the entire prototype chain in order to accommodate usage like the one in the test. See #1496 Reviewed By: fbmal7 Differential Revision: D62357838 fbshipit-source-id: 3873bfd8bcb5c16998dbec67f44ad4c098179758
[LOCAL] Migrate to M1 machines
Recycle block local registers in fast pass (#1448) Summary: Original Author: neildhar@meta.com Original Git: 6b69a06 Original Reviewed By: avp Original Revision: D59072005 The register allocator has the ability to honour a memory limit that is proportional to the product of the number of instructions and basic blocks in the function being allocated. Unfortunately, functions that hit this limit by definition have a lot of instructions Even in the most degenerate case where every block has one instruction, you need 4000 instructions to hit the 10M limit. This diff tries to improve the quality of generated code in cases where most values are used within the basic block they are defined in. In such cases, we currently make the register available after the end of the block. With this diff, the registers become available after their last use in the block. This is useful for functions with extremely large basic blocks, where the current approach would end up allocating a huge number of registers since the registers cannot be used within the same block. Closes #1448 Reviewed By: avp Differential Revision: D60241766 fbshipit-source-id: 5196333862517cd546d675cf8fe005eb1ed5a790
Recycle block local registers in fast pass (#1448) Summary: Original Author: neildhar@meta.com Original Git: 6b69a06 Original Reviewed By: avp Original Revision: D59072005 The register allocator has the ability to honour a memory limit that is proportional to the product of the number of instructions and basic blocks in the function being allocated. Unfortunately, functions that hit this limit by definition have a lot of instructions Even in the most degenerate case where every block has one instruction, you need 4000 instructions to hit the 10M limit. This diff tries to improve the quality of generated code in cases where most values are used within the basic block they are defined in. In such cases, we currently make the register available after the end of the block. With this diff, the registers become available after their last use in the block. This is useful for functions with extremely large basic blocks, where the current approach would end up allocating a huge number of registers since the registers cannot be used within the same block. Closes #1448 Reviewed By: avp Differential Revision: D60241766 fbshipit-source-id: 5196333862517cd546d675cf8fe005eb1ed5a790
PreviousNext