Changes between Version 26 and Version 27 of GcnInstrsVop1


Ignore:
Timestamp:
Jun 16, 2017, 7:00:24 PM (2 years ago)
Author:
trac
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • GcnInstrsVop1

    v26 v27  
    189189NOTE: OMOD and CLAMP modifier affects only for instruction that output is
    190190floating point value.<br />
    191 NOTE: ABS and negation is applied to source operand for any instruction.<br />
    192 NOTE: OMOD modifier doesn't work for half precision (FP16) instructions.</p>
     191NOTE: ABS and negation is applied to source operand for any instruction.  </p>
    193192<p>Negation and absolute value can be combined: <code>-ABS(V0)</code>. Modifiers CLAMP and
    194193OMOD (MUL:2, MUL:4 and DIV:2) can be given in random order.</p>
     
    11371136Description: Convert single FP value to half floating point value with rounding from
    11381137MODE register (single FP rounding mode), and store result to VDST.
    1139 If absolute value is too high, then store -/+infinity to VDST.<br />
     1138If absolute value is too high, then store -/+infinity to VDST.
     1139In GCN 1.2 flushing denormals controlled by MODE. In GCN 1.0/1.1, denormals are enabled.<br />
    11401140Operation:<br />
    11411141<code>VDST = CVTHALF(ASFLOAT(SRC0))</code></p>
     1142<h4>V_CVT_F16_U16</h4>
     1143<p>Opcode: VOP1: 57 (0x39) for GCN 1.2<br />
     1144Opcode VOP3A: 377 (0x179) for GCN 1.2<br />
     1145Syntax: V_CVT_F16_U16 VDST, SRC0<br />
     1146Description: Convert 16-bit unsigned valut to half floating point value.<br />
     1147Operation:<br />
     1148<code>VDST = (HALF)SRC0</code></p>
    11421149<h4>V_CVT_F32_F16</h4>
    11431150<p>Opcode VOP1: 11 (0xb)<br />
     
    11451152Syntax: V_CVT_F32_F16 VDST, SRC0<br />
    11461153Description: Convert half FP value to single FP value, and store result to VDST.
    1147 <strong>By default, immediate is in FP32 format!</strong>.<br />
     1154<strong>By default, immediate is in FP32 format!</strong>.
     1155In GCN 1.2 flushing denormals controlled by MODE. In GCN 1.0/1.1, denormals are enabled.<br />
    11481156Operation:<br />
    11491157<code>VDST = (FLOAT)(ASHALF(SRC0))</code></p>