Refactor: extract byte-identical platform headers/sources to common (-1000 lines)#934
Conversation
|
Warning Review limit reached
More reviews will be available in 3 minutes and 11 seconds. Learn how PR review limits work. Your organization has run out of usage credits. Purchase more in the billing tab. ⌛ How to resolve this issue?After more reviews become available, a review can be triggered using the We recommend that you space out your commits to avoid hitting the rate limit. 🚦 How do rate limits work?CodeRabbit enforces hourly rate limits for each developer per organization. Our paid plans include higher PR review limits than trial, open-source, and free plans. In all cases, reviews become available again over time. During sustained high-volume PR review activity, CodeRabbit may temporarily slow when the next review becomes available. Please see our Fair Usage Limits Policy for further information. ℹ️ Review info⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (36)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
There was a problem hiding this comment.
Code Review
This pull request restructures the codebase by moving shared platform headers and source files (such as logging, scope stats, and compile strategies) from architecture-specific directories into a unified common platform directory (src/common/platform). To accommodate this relocation, include paths and source globs are updated across various CMakeLists.txt files for both a2a3 and a5 platforms, and the Python kernel compiler is adjusted. The review feedback highlights a potential issue with CMake's file(GLOB ...) usage in several CMakeLists.txt files, recommending the addition of the CONFIGURE_DEPENDS flag to ensure the build system automatically detects when source files are added or removed.
2f92daa to
981626a
Compare
Move 6 headers + 1 source whose per-arch differences were ONLY:
- Header guards (SRC_A2A3_… vs SRC_A5_…)
- Comments naming the arch ("a2a3" / "a5" — generalized to the
shared "onboard" / "sim" axis)
- Two log-string tweaks in tensor_dump_aicpu.cpp
into src/common/platform/include/ + src/common/platform/src/aicpu/.
No code-level semantic changes; the per-arch copies were equivalent
modulo doc strings.
Files moved (canonical copy now in common):
- include/common/memory_barrier.h (63)
- include/aicpu/tensor_dump_aicpu.h (305)
- include/aicpu/device_time.h (28)
- include/aicpu/orch_so_file.h (57)
- include/host/function_cache.h (122)
- include/host/platform_compile_info.h (37)
- src/aicpu/tensor_dump_aicpu.cpp (598)
Per-arch saved: 1210
Total dedup'd: ~1100
Comment cleanups carried with the move:
- device_time.h: "a2a3: Real Ascend hardware" → "onboard: …"
- orch_so_file.h: "a2a3 (onboard): pid-based naming" → "onboard: …"
- function_cache.h: "a2a3: Real hardware" → "onboard: …"
- platform_compile_info.h: "Each platform (a2a3, a2a3sim, ...)" →
"Each platform (one of a2a3, a2a3sim, a5, a5sim)"
- Header guards normalized to SRC_COMMON_PLATFORM_INCLUDE_…
Include-path follow-up:
- l2_perf_collector_aicpu.h (still per-arch — DIFF) was doing
`#include "device_time.h"` (no subdirectory prefix), which only
worked because device_time.h lived alongside it in the same
`aicpu/` directory. Now that device_time.h moved up to
common/platform/include/aicpu/, qualify the include as
`#include "aicpu/device_time.h"` to match the shared search-path
shape. Both arches' l2_perf_collector_aicpu.h updated.
Builds:
- a2a3sim + a5sim host_runtime.so clean
- a2a3sim ST L1+L2 pass (38/38, devices 0,1)
- a5sim ST L1+L2 pass (22/22, devices 0,1)
Stacked on hw-native-sys#934 (platform-shared-headers).
Co-authored-by: Chao Wang <26245345+ChaoWao@users.noreply.github.com>
Move 3 more headers whose per-arch differences were ONLY arch-name strings in doc comments — same pattern as hw-native-sys#937: - include/aicore/aicore.h (33 lines) - include/aicpu/device_malloc.h (53 lines) - include/host/memory_allocator.h (109 lines) Per-arch save: ~195 lines deduped (one copy each in common; both arches' copies were equivalent code modulo doc text). Comment cleanup carried with the moves: - aicore.h: "a2a3: Real Ascend hardware" → "onboard: …" and "src/a2a3/platform/onboard/aicore/inner_kernel.h" → "src/{a2a3,a5}/platform/onboard/aicore/inner_kernel.h" - device_malloc.h: "On a2a3:" / "On a2a3sim:" → "On onboard:" / "On sim:" (function descriptions, not behavior) - memory_allocator.h: "a2a3: Wraps CANN runtime" / etc. → "onboard: Wraps CANN runtime" (4 occurrences) - Header guards normalized to SRC_COMMON_PLATFORM_INCLUDE_… Builds: - a2a3sim + a5sim host_runtime.so clean - Smoke ST passes on both arches Stacked on hw-native-sys#937 (which is stacked on hw-native-sys#934). Co-authored-by: Chao Wang <26245345+ChaoWao@users.noreply.github.com>
981626a to
17ff7e8
Compare
Move 6 headers + 1 source whose per-arch differences were ONLY:
- Header guards (SRC_A2A3_… vs SRC_A5_…)
- Comments naming the arch ("a2a3" / "a5" — generalized to the
shared "onboard" / "sim" axis)
- Two log-string tweaks in tensor_dump_aicpu.cpp
into src/common/platform/include/ + src/common/platform/src/aicpu/.
No code-level semantic changes; the per-arch copies were equivalent
modulo doc strings.
Files moved (canonical copy now in common):
- include/common/memory_barrier.h (63)
- include/aicpu/tensor_dump_aicpu.h (305)
- include/aicpu/device_time.h (28)
- include/aicpu/orch_so_file.h (57)
- include/host/function_cache.h (122)
- include/host/platform_compile_info.h (37)
- src/aicpu/tensor_dump_aicpu.cpp (598)
Per-arch saved: 1210
Total dedup'd: ~1100
Comment cleanups carried with the move:
- device_time.h: "a2a3: Real Ascend hardware" → "onboard: …"
- orch_so_file.h: "a2a3 (onboard): pid-based naming" → "onboard: …"
- function_cache.h: "a2a3: Real hardware" → "onboard: …"
- platform_compile_info.h: "Each platform (a2a3, a2a3sim, ...)" →
"Each platform (one of a2a3, a2a3sim, a5, a5sim)"
- Header guards normalized to SRC_COMMON_PLATFORM_INCLUDE_…
Include-path follow-up:
- l2_perf_collector_aicpu.h (still per-arch — DIFF) was doing
`#include "device_time.h"` (no subdirectory prefix), which only
worked because device_time.h lived alongside it in the same
`aicpu/` directory. Now that device_time.h moved up to
common/platform/include/aicpu/, qualify the include as
`#include "aicpu/device_time.h"` to match the shared search-path
shape. Both arches' l2_perf_collector_aicpu.h updated.
Builds:
- a2a3sim + a5sim host_runtime.so clean
- a2a3sim ST L1+L2 pass (38/38, devices 0,1)
- a5sim ST L1+L2 pass (22/22, devices 0,1)
Stacked on hw-native-sys#934 (platform-shared-headers).
Co-authored-by: Chao Wang <26245345+ChaoWao@users.noreply.github.com>
Move 3 more headers whose per-arch differences were ONLY arch-name strings in doc comments — same pattern as hw-native-sys#937: - include/aicore/aicore.h (33 lines) - include/aicpu/device_malloc.h (53 lines) - include/host/memory_allocator.h (109 lines) Per-arch save: ~195 lines deduped (one copy each in common; both arches' copies were equivalent code modulo doc text). Comment cleanup carried with the moves: - aicore.h: "a2a3: Real Ascend hardware" → "onboard: …" and "src/a2a3/platform/onboard/aicore/inner_kernel.h" → "src/{a2a3,a5}/platform/onboard/aicore/inner_kernel.h" - device_malloc.h: "On a2a3:" / "On a2a3sim:" → "On onboard:" / "On sim:" (function descriptions, not behavior) - memory_allocator.h: "a2a3: Wraps CANN runtime" / etc. → "onboard: Wraps CANN runtime" (4 occurrences) - Header guards normalized to SRC_COMMON_PLATFORM_INCLUDE_… Builds: - a2a3sim + a5sim host_runtime.so clean - Smoke ST passes on both arches Stacked on hw-native-sys#937 (which is stacked on hw-native-sys#934). Co-authored-by: Chao Wang <26245345+ChaoWao@users.noreply.github.com>
…-1000 lines)
Move 9 byte-identical headers from src/{a2a3,a5}/platform/include/
into src/common/platform/include/ and 2 byte-identical sources from
src/{a2a3,a5}/platform/src/aicpu/ into src/common/platform/src/aicpu/.
Pure mechanical dedup — no semantic changes.
Files moved (single canonical copy now in common):
- include/aicpu/device_log.h (99)
- include/aicpu/platform_aicpu_affinity.h (19)
- include/aicpu/scope_stats_collector_aicpu.h (90)
- include/common/compile_strategy.h (35)
- include/common/core_type.h (75)
- include/common/qualifier.h (38)
- include/common/scope_stats.h (188)
- include/host/raii_scope_guard.h (56)
- include/host/runtime_compile_info.h (37)
- src/aicpu/scope_stats_collector_aicpu.cpp (331)
- src/aicpu/unified_log_device.cpp (67)
CMakeLists changes:
- 12 platform CMakeLists (a2a3/a5 × onboard/sim × aicore/aicpu/host)
add src/common/platform/include/ to CMAKE_CUSTOM_INCLUDE_DIRS,
ordered after the per-arch include so arch-specific headers
(those that genuinely differ) still take precedence.
- 4 aicpu CMakeLists (a2a3/a5 × onboard/sim) extend the COMMON_SOURCES
glob to pick up the moved .cpp files from common/platform/src/aicpu/.
- simpler_setup/kernel_compiler.py::get_platform_include_dirs() adds
the common include path so orchestration cpp can still resolve
e.g. "common/core_type.h" (used by orchestration_api.h on both
arches).
Onboard + sim, both arches built clean. ST passes:
- a2a3sim L1+L2: 38/38 (devices 0,1)
- a5sim L1+L2: 22/22 (devices 0,1)
Net: -1008 lines deduped across the repo.
Co-authored-by: Chao Wang <26245345+ChaoWao@users.noreply.github.com>
17ff7e8 to
9da6232
Compare
Move 6 headers + 1 source whose per-arch differences were ONLY:
- Header guards (SRC_A2A3_… vs SRC_A5_…)
- Comments naming the arch ("a2a3" / "a5" — generalized to the
shared "onboard" / "sim" axis)
- Two log-string tweaks in tensor_dump_aicpu.cpp
into src/common/platform/include/ + src/common/platform/src/aicpu/.
No code-level semantic changes; the per-arch copies were equivalent
modulo doc strings.
Files moved (canonical copy now in common):
- include/common/memory_barrier.h (63)
- include/aicpu/tensor_dump_aicpu.h (305)
- include/aicpu/device_time.h (28)
- include/aicpu/orch_so_file.h (57)
- include/host/function_cache.h (122)
- include/host/platform_compile_info.h (37)
- src/aicpu/tensor_dump_aicpu.cpp (598)
Per-arch saved: 1210
Total dedup'd: ~1100
Comment cleanups carried with the move:
- device_time.h: "a2a3: Real Ascend hardware" → "onboard: …"
- orch_so_file.h: "a2a3 (onboard): pid-based naming" → "onboard: …"
- function_cache.h: "a2a3: Real hardware" → "onboard: …"
- platform_compile_info.h: "Each platform (a2a3, a2a3sim, ...)" →
"Each platform (one of a2a3, a2a3sim, a5, a5sim)"
- Header guards normalized to SRC_COMMON_PLATFORM_INCLUDE_…
Include-path follow-up:
- l2_perf_collector_aicpu.h (still per-arch — DIFF) was doing
`#include "device_time.h"` (no subdirectory prefix), which only
worked because device_time.h lived alongside it in the same
`aicpu/` directory. Now that device_time.h moved up to
common/platform/include/aicpu/, qualify the include as
`#include "aicpu/device_time.h"` to match the shared search-path
shape. Both arches' l2_perf_collector_aicpu.h updated.
Builds:
- a2a3sim + a5sim host_runtime.so clean
- a2a3sim ST L1+L2 pass (38/38, devices 0,1)
- a5sim ST L1+L2 pass (22/22, devices 0,1)
Stacked on hw-native-sys#934 (platform-shared-headers).
Co-authored-by: Chao Wang <26245345+ChaoWao@users.noreply.github.com>
Move 3 more headers whose per-arch differences were ONLY arch-name strings in doc comments — same pattern as hw-native-sys#937: - include/aicore/aicore.h (33 lines) - include/aicpu/device_malloc.h (53 lines) - include/host/memory_allocator.h (109 lines) Per-arch save: ~195 lines deduped (one copy each in common; both arches' copies were equivalent code modulo doc text). Comment cleanup carried with the moves: - aicore.h: "a2a3: Real Ascend hardware" → "onboard: …" and "src/a2a3/platform/onboard/aicore/inner_kernel.h" → "src/{a2a3,a5}/platform/onboard/aicore/inner_kernel.h" - device_malloc.h: "On a2a3:" / "On a2a3sim:" → "On onboard:" / "On sim:" (function descriptions, not behavior) - memory_allocator.h: "a2a3: Wraps CANN runtime" / etc. → "onboard: Wraps CANN runtime" (4 occurrences) - Header guards normalized to SRC_COMMON_PLATFORM_INCLUDE_… Builds: - a2a3sim + a5sim host_runtime.so clean - Smoke ST passes on both arches Stacked on hw-native-sys#937 (which is stacked on hw-native-sys#934). Co-authored-by: Chao Wang <26245345+ChaoWao@users.noreply.github.com>
….cpp (-250 lines) Move 3 more headers whose per-arch differences were ONLY arch-name strings in doc comments (same pattern as hw-native-sys#937): - include/aicore/aicore.h (33 lines) - include/aicpu/device_malloc.h (53 lines) - include/host/memory_allocator.h (109 lines) Per-arch save from header moves: ~195 lines. Also collapses the 4 per-arch `platform_compile_info.cpp` files (byte-identical modulo the platform-name string literal) into a single shared impl at `src/common/platform/src/host/platform_compile_info.cpp`. The string is supplied via `target_compile_definitions(host_runtime PRIVATE SIMPLER_PLATFORM_NAME=\"…\")` on each per-arch host CMakeLists, matching what `get_platform()` returned before. - 4 files deleted (a2a3/{onboard,sim}/host/, a5/{onboard,sim}/host/) - 1 file added (src/common/platform/src/host/platform_compile_info.cpp) - 4 host CMakeLists: source path moved to the common location + `SIMPLER_PLATFORM_NAME` define added with the appropriate string ("a2a3" / "a2a3sim" / "a5" / "a5sim") Header comment cleanups (same shape as hw-native-sys#937): - `aicore.h`: "a2a3: Real Ascend hardware" → "onboard: …" and doc reference generalized to `src/{a2a3,a5}/platform/...` - `device_malloc.h`: "On a2a3:" / "On a2a3sim:" → "On onboard:" / "On sim:" - `memory_allocator.h`: "a2a3: Wraps CANN runtime" → "onboard: …" - Header guards normalized to SRC_COMMON_PLATFORM_INCLUDE_… Builds: - a2a3sim + a5sim + a2a3 + a5 host_runtime.so clean - `get_platform()` returns the right string per variant (verified via compile_commands.json + ST smoke run) - Smoke ST passes on both arches Stacked on hw-native-sys#937 (which is stacked on hw-native-sys#934). Co-authored-by: Chao Wang <26245345+ChaoWao@users.noreply.github.com>
Move 6 headers + 1 source whose per-arch differences were ONLY:
- Header guards (SRC_A2A3_… vs SRC_A5_…)
- Comments naming the arch ("a2a3" / "a5" — generalized to the
shared "onboard" / "sim" axis)
- Two log-string tweaks in tensor_dump_aicpu.cpp
into src/common/platform/include/ + src/common/platform/src/aicpu/.
No code-level semantic changes; the per-arch copies were equivalent
modulo doc strings.
Files moved (canonical copy now in common):
- include/common/memory_barrier.h (63)
- include/aicpu/tensor_dump_aicpu.h (305)
- include/aicpu/device_time.h (28)
- include/aicpu/orch_so_file.h (57)
- include/host/function_cache.h (122)
- include/host/platform_compile_info.h (37)
- src/aicpu/tensor_dump_aicpu.cpp (598)
Per-arch saved: 1210
Total dedup'd: ~1100
Comment cleanups carried with the move:
- device_time.h: "a2a3: Real Ascend hardware" → "onboard: …"
- orch_so_file.h: "a2a3 (onboard): pid-based naming" → "onboard: …"
- function_cache.h: "a2a3: Real hardware" → "onboard: …"
- platform_compile_info.h: "Each platform (a2a3, a2a3sim, ...)" →
"Each platform (one of a2a3, a2a3sim, a5, a5sim)"
- Header guards normalized to SRC_COMMON_PLATFORM_INCLUDE_…
Include-path follow-up:
- l2_perf_collector_aicpu.h (still per-arch — DIFF) was doing
`#include "device_time.h"` (no subdirectory prefix), which only
worked because device_time.h lived alongside it in the same
`aicpu/` directory. Now that device_time.h moved up to
common/platform/include/aicpu/, qualify the include as
`#include "aicpu/device_time.h"` to match the shared search-path
shape. Both arches' l2_perf_collector_aicpu.h updated.
Builds:
- a2a3sim + a5sim host_runtime.so clean
- a2a3sim ST L1+L2 pass (38/38, devices 0,1)
- a5sim ST L1+L2 pass (22/22, devices 0,1)
Stacked on hw-native-sys#934 (platform-shared-headers).
Co-authored-by: Chao Wang <26245345+ChaoWao@users.noreply.github.com>
Move 6 headers + 1 source whose per-arch differences were ONLY:
- Header guards (SRC_A2A3_… vs SRC_A5_…)
- Comments naming the arch ("a2a3" / "a5" — generalized to the
shared "onboard" / "sim" axis)
- Two log-string tweaks in tensor_dump_aicpu.cpp
into src/common/platform/include/ + src/common/platform/src/aicpu/.
No code-level semantic changes; the per-arch copies were equivalent
modulo doc strings.
Files moved (canonical copy now in common):
- include/common/memory_barrier.h (63)
- include/aicpu/tensor_dump_aicpu.h (305)
- include/aicpu/device_time.h (28)
- include/aicpu/orch_so_file.h (57)
- include/host/function_cache.h (122)
- include/host/platform_compile_info.h (37)
- src/aicpu/tensor_dump_aicpu.cpp (598)
Per-arch saved: 1210
Total dedup'd: ~1100
Comment cleanups carried with the move:
- device_time.h: "a2a3: Real Ascend hardware" → "onboard: …"
- orch_so_file.h: "a2a3 (onboard): pid-based naming" → "onboard: …"
- function_cache.h: "a2a3: Real hardware" → "onboard: …"
- platform_compile_info.h: "Each platform (a2a3, a2a3sim, ...)" →
"Each platform (one of a2a3, a2a3sim, a5, a5sim)"
- Header guards normalized to SRC_COMMON_PLATFORM_INCLUDE_…
Include-path follow-up:
- l2_perf_collector_aicpu.h (still per-arch — DIFF) was doing
`#include "device_time.h"` (no subdirectory prefix), which only
worked because device_time.h lived alongside it in the same
`aicpu/` directory. Now that device_time.h moved up to
common/platform/include/aicpu/, qualify the include as
`#include "aicpu/device_time.h"` to match the shared search-path
shape. Both arches' l2_perf_collector_aicpu.h updated.
Builds:
- a2a3sim + a5sim host_runtime.so clean
- a2a3sim ST L1+L2 pass (38/38, devices 0,1)
- a5sim ST L1+L2 pass (22/22, devices 0,1)
Stacked on #934 (platform-shared-headers).
Co-authored-by: Chao Wang <26245345+ChaoWao@users.noreply.github.com>
….cpp (-250 lines) Move 3 more headers whose per-arch differences were ONLY arch-name strings in doc comments (same pattern as hw-native-sys#937): - include/aicore/aicore.h (33 lines) - include/aicpu/device_malloc.h (53 lines) - include/host/memory_allocator.h (109 lines) Per-arch save from header moves: ~195 lines. Also collapses the 4 per-arch `platform_compile_info.cpp` files (byte-identical modulo the platform-name string literal) into a single shared impl at `src/common/platform/src/host/platform_compile_info.cpp`. The string is supplied via `target_compile_definitions(host_runtime PRIVATE SIMPLER_PLATFORM_NAME=\"…\")` on each per-arch host CMakeLists, matching what `get_platform()` returned before. - 4 files deleted (a2a3/{onboard,sim}/host/, a5/{onboard,sim}/host/) - 1 file added (src/common/platform/src/host/platform_compile_info.cpp) - 4 host CMakeLists: source path moved to the common location + `SIMPLER_PLATFORM_NAME` define added with the appropriate string ("a2a3" / "a2a3sim" / "a5" / "a5sim") Header comment cleanups (same shape as hw-native-sys#937): - `aicore.h`: "a2a3: Real Ascend hardware" → "onboard: …" and doc reference generalized to `src/{a2a3,a5}/platform/...` - `device_malloc.h`: "On a2a3:" / "On a2a3sim:" → "On onboard:" / "On sim:" - `memory_allocator.h`: "a2a3: Wraps CANN runtime" → "onboard: …" - Header guards normalized to SRC_COMMON_PLATFORM_INCLUDE_… Builds: - a2a3sim + a5sim + a2a3 + a5 host_runtime.so clean - `get_platform()` returns the right string per variant (verified via compile_commands.json + ST smoke run) - Smoke ST passes on both arches Stacked on hw-native-sys#937 (which is stacked on hw-native-sys#934). Co-authored-by: Chao Wang <26245345+ChaoWao@users.noreply.github.com>
….cpp (-250 lines) (#938) Move 3 more headers whose per-arch differences were ONLY arch-name strings in doc comments (same pattern as #937): - include/aicore/aicore.h (33 lines) - include/aicpu/device_malloc.h (53 lines) - include/host/memory_allocator.h (109 lines) Per-arch save from header moves: ~195 lines. Also collapses the 4 per-arch `platform_compile_info.cpp` files (byte-identical modulo the platform-name string literal) into a single shared impl at `src/common/platform/src/host/platform_compile_info.cpp`. The string is supplied via `target_compile_definitions(host_runtime PRIVATE SIMPLER_PLATFORM_NAME=\"…\")` on each per-arch host CMakeLists, matching what `get_platform()` returned before. - 4 files deleted (a2a3/{onboard,sim}/host/, a5/{onboard,sim}/host/) - 1 file added (src/common/platform/src/host/platform_compile_info.cpp) - 4 host CMakeLists: source path moved to the common location + `SIMPLER_PLATFORM_NAME` define added with the appropriate string ("a2a3" / "a2a3sim" / "a5" / "a5sim") Header comment cleanups (same shape as #937): - `aicore.h`: "a2a3: Real Ascend hardware" → "onboard: …" and doc reference generalized to `src/{a2a3,a5}/platform/...` - `device_malloc.h`: "On a2a3:" / "On a2a3sim:" → "On onboard:" / "On sim:" - `memory_allocator.h`: "a2a3: Wraps CANN runtime" → "onboard: …" - Header guards normalized to SRC_COMMON_PLATFORM_INCLUDE_… Builds: - a2a3sim + a5sim + a2a3 + a5 host_runtime.so clean - `get_platform()` returns the right string per variant (verified via compile_commands.json + ST smoke run) - Smoke ST passes on both arches Stacked on #937 (which is stacked on #934). Co-authored-by: Chao Wang <26245345+ChaoWao@users.noreply.github.com>
Summary
Move 9 byte-identical headers from
src/{a2a3,a5}/platform/include/intosrc/common/platform/include/and 2 byte-identical sources fromsrc/{a2a3,a5}/platform/src/aicpu/intosrc/common/platform/src/aicpu/. Pure mechanical dedup — no semantic changes.Files moved (single canonical copy now in common)
include/aicpu/device_log.hinclude/aicpu/platform_aicpu_affinity.hinclude/aicpu/scope_stats_collector_aicpu.hinclude/common/compile_strategy.hinclude/common/core_type.hinclude/common/qualifier.hinclude/common/scope_stats.hinclude/host/raii_scope_guard.hinclude/host/runtime_compile_info.hsrc/aicpu/scope_stats_collector_aicpu.cppsrc/aicpu/unified_log_device.cppCMakeLists changes
src/common/platform/include/toCMAKE_CUSTOM_INCLUDE_DIRS, ordered after the per-arch include so arch-specific headers (those that genuinely differ) still take precedence.COMMON_SOURCESglob to pick up the moved .cpp files fromcommon/platform/src/aicpu/.simpler_setup/kernel_compiler.py::get_platform_include_dirs()adds the common include path so orchestration cpp can still resolve e.g.\"common/core_type.h\"(used byorchestration_api.hon both arches).Test plan
bgemm(host_build_graph) on a2a3sim — was the first to catch the orchestration include-path miss; passes after the kernel_compiler.py fixNet: -1008 lines deduped across the repo.