Searched refs:SDWA (Results 1 - 14 of 14) sorted by relevance

/xsrc/external/mit/MesaLib/dist/docs/relnotes/
H A D21.3.4.rst70 - aco: don't allow SDWA on VOP3P instructions
H A D20.1.1.rst157 - aco: consider SDWA during value numbering
H A D21.3.0.rst990 - aco: add more validation rules for SDWA operands
1006 - aco: rewrite SDWA selector
1008 - aco/print_ir: always print SDWA dst & src selections
4178 - aco/optimizer: Skip SDWA on v_lshlrev when unnecessary in apply_extract.
H A D21.2.0.rst1571 - aco: refactor SDWA opcode validation
4399 - aco: allow SDWA sels smaller than the operand size
4440 - aco: disallow SDWA for instructions with 64-bit definitions/operands
4446 - aco: optimize 32-bit extracts and inserts using SDWA
4450 - aco/tests: add SDWA tests
H A D20.2.0.rst3811 - aco: consider SDWA during value numbering
3962 - aco: don't apply constant to SDWA on GFX8
H A D20.3.0.rst4035 - aco: don't apply constant to SDWA on GFX8
4113 - aco: handle SDWA in the optimizer
H A D21.1.0.rst4836 - aco: allow SDWA sels smaller than the operand size
/xsrc/external/mit/MesaLib/dist/src/amd/compiler/
H A Daco_ir.h64 * - VOP2* | SDWA represents a VOP2 instruction with sub-dword addressing.
106 SDWA = 1 << 14, enumerator in enum:aco::Format
291 return (Format)((uint32_t)Format::SDWA | (uint32_t)format);
515 * constant can be implemented with a SDWA v_mul_u32_u24. So consider all
1306 constexpr bool isSDWA() const noexcept { return (uint16_t)format & (uint16_t)Format::SDWA; }
H A Daco_ir.cpp258 (Format)(((uint16_t)tmp->format & ~(uint16_t)Format::VOP3) | (uint16_t)Format::SDWA);
275 /* SDWA only uses operands 0 and 1. */
H A Daco_assembler.cpp694 /* first emit the instruction without the SDWA operand */
697 instr->format = (Format)((uint16_t)instr->format & ~(uint16_t)Format::SDWA);
H A Daco_validate.cpp118 base_format = (Format)((uint32_t)base_format & ~(uint32_t)Format::SDWA);
149 /* check SDWA */
153 "Format cannot have SDWA applied", instr.get());
155 check(program->chip_class >= GFX8, "SDWA is GFX8+ only", instr.get());
159 "SDWA omod only supported on GFX9+", instr.get());
162 "SDWA VOPC clamp only supported on GFX8", instr.get());
165 "SDWA+VOPC definition must be fixed to vcc on GFX8", instr.get());
168 check(def.bytes() <= 4, "SDWA definitions must not be larger than 4 bytes",
171 "SDWA definition selection size must be at most definition size", instr.get());
174 "SDWA definitio
[all...]
H A Daco_opcodes.py78 SDWA = 1 << 14 variable in class:Format
179 if self == Format.SDWA:
H A Daco_optimizer.cpp1532 if (instr->format == Format::VOPC && /* don't optimize VOP3 / SDWA / DPP */
2090 new_opcode, (Format)((uint16_t)Format::SDWA | (uint16_t)Format::VOPC), 2, 1);
2317 /* TODO: this can handle SDWA min/max instructions by using opsel */
2826 bool can_use_omod = (can_vop3 || ctx.program->chip_class >= GFX9); /* SDWA omod is GFX9+ */
3244 /* Apply SDWA. Do this after label_instruction() so it can remove
3245 * label_extract if not all instructions can take SDWA. */
/xsrc/external/mit/MesaLib/dist/
H A D.pick_status.json18373 "description": "aco: don't allow SDWA on VOP3P instructions",
[all...]

Completed in 154 milliseconds