Intel® C++ Compiler 16.0 User and Reference Guide
Gathers 2/4 packed single-precision floating point values from memory referenced by the given base address, qword indices and scale, and using the given single-precision FP mask values. The corresponding Intel® AVX2 instruction is VGATHERQPS.
extern __m128 _mm_mask_i64gather_ps(__m128 def_vals, float const * base, __m128i vindex, __m128 vmask, const int scale); |
extern __m128 _mm256_mask_i64gather_ps(float const * base, __m256i vindex, __m256i vmask, const int scale); |
def_vals |
the vector of single-precision FP values copied to the destination when the corresponding element of the single-precision FP mask is '0'. |
base |
the base address used to reference the loaded FP elements. |
vindex |
the vector of qword indices used to reference the loaded FP elements. |
vmask |
the vector of FP elements used as a vector mask; only the most significant bit of each data element is used as a mask. |
scale |
32-bit scale used to address the loaded FP elements. |
The intrinsics conditionally load 2/4 packed single-precision floating-point values from memory using qword indices and updates the destination operand. The intrinsic _mm_mask_i64gather_ps() also sets the upper 64-bits of the result to '0'.
Below is the pseudo-code for the intrinsics:
_mm_mask_i64gather_ps():
result[31:0] = (vmask[31]==1) ? (mem[base+vindex[63:0]*scale]) : (def_vals[31:0]); result[63:32] = (vmask[63]==1) ? (mem[base+vindex[127:64]*scale]) : (def_vals[63:32]); result[127:64] = 0;
_mm256_mask_i64gather_ps():
result[31:0] = (vmask[31]==1) ? (mem[base+vindex[63:0]*scale]) : (def_vals[31:0]); result[63:32] = (vmask[63]==1) ? (mem[base+vindex[127:64]*scale]) : (def_vals[63:32]); result[95:64] = (vmask[95]==1) ? (mem[base+vindex[191:128]*scale]) : (def_vals[95:64]); result[127:96] = (vmask[127]==1) ? (mem[base+vindex[255:192]*scale]) : (def_vals[127:96]);
A 256/128-bit vector with conditionally gathered single-precision FP values.