llvm.org GIT mirror llvm / master docs / AMDGPU / gfx8_dst_mimg_regular_d16.rst
master

Tree @master (Download .tar.gz)

gfx8_dst_mimg_regular_d16.rst @masterview markup · raw · history · blame

vdst

Image data to load by an image instruction.

Size: depends on :ref:`dmask<amdgpu_synid_dmask>`, :ref:`tfe<amdgpu_synid_tfe>` and :ref:`d16<amdgpu_synid_d16>`:

  • :ref:`dmask<amdgpu_synid_dmask>` may specify from 1 to 4 data elements. Each data element occupies either 32 bits or 16 bits depending on :ref:`d16<amdgpu_synid_d16>`.
  • :ref:`d16<amdgpu_synid_d16>` has different meaning for GFX8.0 and GFX8.1:
    • For GFX8.0 this modifier does not affect size of data elements in registers. Data in registers are stored in low 16 bits, high 16 bits are unused. There is no packing.
    • Starting from GFX8.1 this modifier specifies that data elements in registers are packed; each value occupies 16 bits.
  • :ref:`tfe<amdgpu_synid_tfe>` adds 1 dword if specified.

Operands: :ref:`v<amdgpu_synid_v>`