The following intrinsics perform faster floating point calculations using the 32-bit (FPU32) and 64-bit (FPU64) hardware floating-point support. These intrinsics are enabled if the --float_support compiler option is set to fpu32 or fpu64, respectively.
If you are using the COFF ABI, arguments and return values listed as floats may also use doubles, because both are 32-bit types. The FPU32 intrinsics are supported, but the FPU64 intrinsics are not supported if you are using the COFF ABI.
If you are using EABI, the float type is 32-bit and the double type is 64-bit. If FPU32 is enabled, use only the FPU32 versions. In general, if FPU64 is enabled, you may use both FPU32 and FPU64 intrinsics. However, in EABI mode with FPU64, there is no instruction that swaps two 32-bit floats. In this situation, you cannot use the __swapf() intrinsic, and the __swapff() intrinsic must only be used with 64-bit doubles.
| FPU Version | Intrinsic | Assembly Instruction(s) | Description |
|---|---|---|---|
| FPU32 | float __einvf32( float x); | EINVF32x | Compute and return 1/x to about 8 bits of precision. |
| FPU64 | double __einvf64( double x); | EINVF64x | Compute and return 1/x to about 8 bits of precision. |
| FPU32 | float __eisqrtf32( float x); | EISQRTF32x | Find the square root of 1/x to about 8 bits of precision. |
| FPU64 | double __eisqrtf64( double x); | EISQRTF64x | Find the square root of 1/x to about 8 bits of precision. |
| FPU32 | void __f32_max_idx( float &dst, float src,
float &idx_dst, float idx_src); |
MAXF32dst, src
|| MOV32idx_dst, idx_src |
If src>dst, copy src to dst, and copy idx_src to idx_dst. |
| FPU64 | void __f64_max_idx( double &dst, double src,
double &idx_dst, double idx_src); |
MAXF64dst, src
|| MOV64idx_dst, idx_src |
If src>dst, copy src to dst, and copy idx_src to idx_dst. |
| FPU32 | void __f32_min_idx( float &dst, float src,
float &idx_dst, float idx_src); |
MINF32dst, src
|| MOV32idx_dst, idx_src |
If src<dst, copy src to dst, and copy idx_src to idx_dst. |
| FPU64 | void __f64_min_idx( double &dst, double src,
double &idx_dst, double idx_src); |
MINF64dst, src
|| MOV64idx_dst, idx_src |
If src<dst, copy src to dst, and copy idx_src to idx_dst. |
| FPU32 | int __f32toi16r(float src); | F32TOI16Rdst,src | Convert float to int and round. |
| FPU32 | unsigned int __f32toui16r(float src); | F32TOUI16Rdst,src | Convert float to unsigned int and round. |
| FPU32 | float __fmax( float x, float y); | MAXF32dst, src | If src>dst, copy src to dst. |
| FPU64 | double __fmax64( double x, double y); | MAXF64dst, src | If src>dst, copy src to dst. |
| FPU32 | float __fmin( float x, float y); | MINF32dst, src | If src<dst, copy src to dst. |
| FPU64 | double __fmin64( double x, double y); | MINF64dst, src | If src<dst, copy src to dst. |
| FPU32 | float __fracf32(float src); | FRACF32dst,src | Return the fractional portion of src. |
| FPU64 | double __fracf64(double src); | FRACF64dst,src | Return the fractional portion of src. |
| FPU32 | float __fsat(float val, float max, float min); | MAXF32dst,src2
MINF32dst, src1 |
Return val if min < val < max. Else return min if value < min. Else return max if val > max. |
| FPU32 (only) | void __swapf( float &a, float &b); | swapfa, b | Swap the contents of a and b. |
| FPU32 (only) | void __swapff( float &a, float &b); | swapffa, b | Swap the contents of a and b. |
| FPU64 | void __swapff( double &a, double &b); | swapffa, b | Swap the contents of a and b. |