Changes between Version 26 and Version 27 of GcnInstrsVop3

Timestamp:

06/17/17 17:00:29 (7 years ago)

Author:

trac

Comment:

--

GcnInstrsVop3

@@ -1301 +1301 @@
 Operation:<br />
 <code>VDST = ASFLOAT(SRC0) * ASFLOAT(SRC1) + ASFLOAT(SRC2)</code></p>
+<h4>V_MAD_I16</h4>
+<p>Opcode: 492 (0x1ec) for GCN 1.2<br />
+Syntax: V_MAD_I16 VDST, SRC0, SRC1, SRC2<br />
+Description: Multiply 16-bit signed value from SRC0 by 16-bit signed value from
+SRC1 and add 16-bit signed value from SRC2, and store 16-bit signed result to VDST.<br />
+Operation:<br />
+<code>VDST = (INT16)((INT16)SRC0*(INT16)SRC1 + (INT16)SRC2)</code></p>
 <h4>V_MAD_I32_I24</h4>
 <p>Opcode: 322 (0x142) for GCN 1.0/1.1; 450 (0x1c2) for GCN 1.2<br />
@@ -1332 +1339 @@
 <code>if (ASFLOAT(SRC0)!=0.0 &amp;&amp; ASFLOAT(SRC1)!=0.0)
     VDST = ASFLOAT(SRC0) * ASFLOAT(SRC1) + ASFLOAT(SRC2)</code></p>
+<h4>V_MAD_U16</h4>
+<p>Opcode: 491 (0x1eb) for GCN 1.2<br />
+Syntax: V_MAD_U16 VDST, SRC0, SRC1, SRC2<br />
+Description: Multiply 16-bit unsigned value from SRC0 by 16-bit unsigned value from
+SRC1 and add 16-bit unsigned value from SRC2, and store 16-bit unsigned result to VDST.<br />
+Operation:<br />
+<code>VDST = ((UINT16)SRC0*(UINT16)SRC1 + (UINT16)SRC2) &amp; 0xffff</code></p>
 <h4>V_MAD_U32_U24</h4>
 <p>Opcode: 323 (0x143) for GCN 1.0/1.1; 451 (0x1c3) for GCN 1.2<br />
@@ -1604 +1618 @@
         VDST = ASFLOAT(SRC0) * ASFLOAT(SRC1)
 }</code></p>
+<h4>V_PERM_B32</h4>
+<p>Opcode: 493 (0x1ed) for GCN 1.2<br />
+Syntax: V_PERM_B32 VDST, SRC0, SRC1, SRC2<br />
+Description: Permute bytes. Choose for every byte in dword, specified value. Bytes in
+SRC2 dword selects value for result dword. Value 0-7 choose byte of this index of quadword
+(64-bit value) built from SRC0 (higher bits) and SRC1 (lower bits). Value from 8-11
+choose 0xff<em>BIT, where BIT is last bit from 2</em>N+1 from 64-bit value (SRC0,SRC1).
+Value 12 choose zero. Value equal or greater than 13 choose 0xff.<br />
+Operation:<br />
+<code>VDST = 0
+UINT64 qword = (((UINT64)SRC0)&lt;&lt;32) | SRC1 
+for (int i = 0; i &lt; 4; i++)
+{
+    BYTE choice = (SRC2 &gt;&gt; (8*i)) &amp; 0xff
+    BYTE result
+    if (choice &gt;= 13)
+        result = 0xff
+    else if (choice == 12)
+        result = 0
+    else if (choice &gt;= 8)
+        result = 0xff * qword&gt;&gt;((choice-8)*16 + 15)
+    else
+        result = (qword &gt;&gt; (choice*8)) &amp; 0xff
+    VDST |= (result &lt;&lt; (i*8))
+}</code></p>
 <h4>V_QSAD_U8, V_QSAD_PK_U16_U8</h4>
 <p>Opcode: 370 (0x172) for GCN 1.0/1.1; 485 (0x1e5) for GCN 1.2<br />