Changeset 1762 in CLRX


Ignore:
Timestamp:
Nov 28, 2015, 11:49:24 PM (5 years ago)
Author:
matszpk
Message:

CLRadeonExtender : Updated VOP1 instruction list: Added V_RCP*_F64 instructions, fixed some descriptions.

File:
1 edited

Legend:

Unmodified
Added
Removed
  • CLRadeonExtender/trunk/doc/GcnInstrsVop1.md

    r1761 r1762  
    614614Syntax: V_RCP_CLAMP_F32 VDST, SRC0 
    615615Description: Approximate reciprocal from floating point value SRC0 and store it to VDST.
    616 Guaranted error below 1ulp. Result is clamped maximum float value including its sign
     616Guaranted error below 1ulp. Result is clamped to MAX_FLOAT including sign of a result
    617617Description: 
    618618```
     
    632632```
    633633VDST = APPROX_RCP(ASFLOAT(SRC0))
     634```
     635
     636#### V_RCP_F64
     637
     638Opcode VOP1: 47 (0x2f) for GCN 1.0/1.1; 37 (0x25) for GCN 2.0 
     639Opcode VOP3A: 431 (0x1af) for GCN 1.0/1.1; 357 (0x165) for GCN 2.0 
     640Syntax: V_RCP_F64 VDST(2), SRC0(2) 
     641Description: Approximate reciprocal from double FP value SRC0 and store it to VDST.
     642Relative error of approximation is ~1e-8. 
     643Description: 
     644```
     645VDST = APPROX_RCP(ASDOUBLE(SRC0))
     646```
     647
     648#### V_RCP_CLAMP_F64
     649
     650Opcode VOP1: 48 (0x30) for GCN 1.0/1.1 
     651Opcode VOP3A: 432 (0x1b0) for GCN 1.0/1.1 
     652Syntax: V_RCP_CLAMP_F64 VDST(2), SRC0(2) 
     653Description: Approximate reciprocal from double FP value SRC0 and store it to VDST.
     654Relative error of approximation is ~1e-8.
     655Result is clamped to MAX_DOUBLE value including sign of a result. 
     656Description: 
     657```
     658VDST = APPROX_RCP(ASDOUBLE(SRC0))
     659if (ABS(ASDOUBLE(VDST))==INF)
     660    VDST = SIGN(ASDOUBLE(VDST)) * MAX_DOUBLE
    634661```
    635662
     
    712739```
    713740
     741#### V_RSQ_F32
     742
     743Opcode VOP1: 46 (0x2e) for GCN 1.0/1.1; 36 (0x24) for GCN 2.0 
     744Opcode VOP3A: 430 (0x1ae) for GCN 1.0/1.1; 356 (0x164) for GCN 2.0 
     745Syntax: V_RCP_F32 VDST, SRC0 
     746Description: Approximate reciprocal square root from floating point value SRC0 and
     747store it to VDST. If SRC0 is negative value, store -NAN to VDST.
     748This instruction doesn't handle denormalized values regardless FLOAT MODE register setup. 
     749Description: 
     750```
     751VDST = APPROX_RSQRT(ASFLOAT(SRC0))
     752```
     753
    714754#### V_RSQ_LEGACY_F32
    715755
     
    728768```
    729769
    730 #### V_RSQ_F32
    731 
    732 Opcode VOP1: 46 (0x2e) for GCN 1.0/1.1; 36 (0x24) for GCN 2.0 
    733 Opcode VOP3A: 430 (0x1ae) for GCN 1.0/1.1; 356 (0x164) for GCN 2.0 
    734 Syntax: V_RCP_F32 VDST, SRC0 
    735 Description: Approximate reciprocal square root from floating point value SRC0 and
    736 store it to VDST. If SRC0 is negative value, store -NAN to VDST.
    737 This instruction doesn't handle denormalized values regardless FLOAT MODE register setup. 
    738 Description: 
    739 ```
    740 VDST = APPROX_RSQRT(ASFLOAT(SRC0))
    741 ```
    742 
    743770#### V_TRUNC_F32
    744771
Note: See TracChangeset for help on using the changeset viewer.