You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: README_DEVELOPERS.md
+5Lines changed: 5 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -80,6 +80,7 @@ The main [README.md](README.md) keeps the simplest supported build path. This se
80
80
-`-DMINIEXPR_BUILD_EXAMPLES=ON|OFF`
81
81
-`-DMINIEXPR_BUILD_BENCH=ON|OFF`
82
82
-`-DMINIEXPR_USE_SLEEF=ON|OFF`
83
+
-`-DMINIEXPR_USE_ACCELERATE=ON|OFF` (macOS only)
83
84
-`-DMINIEXPR_ENABLE_TCC_JIT=ON|OFF`
84
85
-`-DMINIEXPR_BUILD_BUNDLED_LIBTCC=ON|OFF` (build bundled libtcc from minicc when TCC JIT is enabled)
85
86
-`-DMINIEXPR_DSL_TRACE_DEFAULT=ON|OFF` (emit DSL trace logs by default when `ME_DSL_TRACE` is unset)
@@ -91,6 +92,8 @@ The main [README.md](README.md) keeps the simplest supported build path. This se
91
92
- On Emscripten, setting `MINIEXPR_ENABLE_TCC_JIT=ON` enables wasm32 JIT support automatically.
92
93
- Setting `MINIEXPR_ENABLE_TCC_JIT=OFF` disables TCC-based JIT backends; on Linux/macOS, the separate `# me:compiler=cc` runtime path may still be available.
93
94
-`MINIEXPR_USE_SLEEF=ON` fetches SLEEF and enables SIMD math acceleration; set it to `OFF` to build without SLEEF.
95
+
-`MINIEXPR_USE_ACCELERATE=ON` enables the macOS Accelerate/vForce backend; in `auto` mode on macOS it is preferred by default, and unsupported functions still fall back to scalar kernels.
96
+
- When `ME_SIMD_MATH_BACKEND=accelerate` is active, the `ME_SIMD_ULP_1` / `ME_SIMD_ULP_3_5` distinction does not select different kernels. Those accuracy modes remain meaningful for the SLEEF backend.
94
97
95
98
### Alternative Build Invocations
96
99
@@ -118,6 +121,8 @@ The public/runtime-stable DSL JIT controls remain documented in [README.md](READ
118
121
119
122
### Internal/Test-Only Environment Variables
120
123
124
+
-`ME_SIMD_MATH_BACKEND=auto|sleef|accelerate|scalar`: Force the SIMD math backend selection used by `src/functions-simd.c` for benchmarking and debugging. Default: `auto` (`accelerate` on macOS when enabled, otherwise the existing platform backend selection).
125
+
- The SIMD math benchmarks print backend-aware columns. For `accelerate` and `scalar`, do not interpret the `ME_SIMD_ULP_1` / `ME_SIMD_ULP_3_5` labels as distinct math implementations.
121
126
-`ME_DSL_WHILE_MAX_ITERS=<n>`: Override the runtime safety cap for DSL `while` loops.
122
127
-`ME_DSL_JIT_MATH_BRIDGE=0|1`: Enable or disable runtime math-bridge lowering globally. Default: `1`.
123
128
-`ME_DSL_JIT_SCALAR_MATH_BRIDGE=0|1`: Enable scalar math-bridge lowering for the `cc` backend. Default: `0`.
0 commit comments