llvm.org GIT mirror llvm / release_80 docs / AMDGPU / gfx8_dst_buf_d16_96.rst

Tree @release_80 (Download .tar.gz)

gfx8_dst_buf_d16_96.rst @release_80view markup · raw · history · blame


Instruction output: data read from a memory buffer and converted to a 16-bit format.

Size: depends on GFX8 GPU revision and :ref:`tfe<amdgpu_synid_tfe>`:

  • 3 dwords for GFX8.0. This H/W supports no packing.
  • 2 dwords for GFX8.1+. This H/W supports data packing.
  • :ref:`tfe<amdgpu_synid_tfe>` adds one dword if specified.

Operands: :ref:`v<amdgpu_synid_v>`