IP PortfolioDatapath and DSPfloating point
of_mx_quant
Microscaling (MX) block quantizer
Maturity plannedTargets FPGA + ASICDifficulty MediumPriority 2
Converts FP32/bfloat16 tensors to OCP Microscaling formats (MXFP8, MXFP6, MXFP4, MXINT8) with a shared 8-bit block exponent per 32 elements, and back.
Planned. This block is specified in the catalog with its interfaces and targets. No implementation exists yet; the skeleton command below generates a lint-clean shell with its ports and registers.
Interfaces
| Port | Direction | Kind | Width | Notes |
|---|---|---|---|---|
| s_in | in | AXI4-Stream | 512 | |
| m_out | out | AXI4-Stream | 264 |
Standards
- OCP Microscaling Formats (MX) v1.0
More in floating point
| ID | Block | Maturity |
|---|---|---|
| of_fp_add | Floating-point adder/subtractor | simulated |
| of_fp_mul | Floating-point multiplier | simulated |
| of_fp_round | Floating-point round & pack unit | simulated |
| of_fp2int | Floating-point to fixed-point converter | simulated |
| of_int2fp | Fixed-point to floating-point converter | simulated |
| of_fp_fma | Floating-point fused multiply-add | simulated |
| of_fp_divsqrt | Floating-point divide and square root | simulated |
| of_fp_cmp | Floating-point compare, min/max and classify | simulated |
| of_fp_cvt | Floating-point format converter | simulated |
| of_bf16_dotp | bfloat16 dot-product unit | simulated |