Changes between Version 5 and Version 6 of GcnInstrsVop3

Timestamp:

12/06/15 21:00:18 (8 years ago)

Author:

trac

Comment:

--

GcnInstrsVop3

@@ -1110 +1110 @@
         VDST = ASFLOAT(SRC0) * ASFLOAT(SRC1)
 }</code></p>
+<h4>V_SAD_HI_U8</h4>
+<p>Opcode: 347 (0x15b) for GCN 1.0/1.1; 474 (0x1da) for GCN 1.2<br />
+Syntax: V_SAD_HI_U8 VDST, SRC0, SRC1, SRC2<br />
+Description: Calculate sum of absolute differences for all four bytes in SRC0 and SRC1,
+shift result to high 16-bits, add SRC2 to result, and store result to VDST.<br />
+Operation:<br />
+<code>VDST = SRC2
+for (UINT8 i = 0; i &lt; 4; i++)
+    VDST += (ABS(((SRC0 &gt;&gt; (i*8)) &amp; 0xff) - ((SRC1 &gt;&gt; (i*8)) &amp; 0xff)))&lt;&lt;16</code></p>
+<h4>V_SAD_U16</h4>
+<p>Opcode: 348 (0x15c) for GCN 1.0/1.1; 475 (0x1db) for GCN 1.2<br />
+Syntax: V_SAD_U16 VDST, SRC0, SRC1, SRC2<br />
+Description: Calculate sum of absolute differences for two 16-bit words in SRC0 and SRC1, add
+SRC2 to result, and store result to VDST.<br />
+Operation:<br />
+<code>VDST = SRC2
+VDST += ABS((SRC0 &amp; 0xffff) - (SRC1 &amp; 0xffff))
+VDST += ABS((SRC0 &gt;&gt; 16) - (SRC1 &gt;&gt; 16))</code></p>
+<h4>V_SAD_U32</h4>
+<p>Opcode: 349 (0x15d) for GCN 1.0/1.1; 476 (0x1dc) for GCN 1.2<br />
+Syntax: V_SAD_U32 VDST, SRC0, SRC1, SRC2<br />
+Description: Calculate sum of absolute difference for SRC0 and SRC1, add
+SRC2 to result, and store result to VDST.<br />
+Operation:<br />
+<code>VDST = SRC2 + ABS(SRC0 - SRC1)</code></p>
+<h4>V_SAD_U8</h4>
+<p>Opcode: 346 (0x15a) for GCN 1.0/1.1; 473 (0x1d9) for GCN 1.2<br />
+Syntax: V_SAD_U8 VDST, SRC0, SRC1, SRC2<br />
+Description: Calculate sum of absolute differences for all four bytes in SRC0 and SRC1, add
+SRC2 to result, and store result to VDST.<br />
+Operation:<br />
+<code>VDST = SRC2
+for (UINT8 i = 0; i &lt; 4; i++)
+    VDST += ABS(((SRC0 &gt;&gt; (i*8)) &amp; 0xff) - ((SRC1 &gt;&gt; (i*8)) &amp; 0xff))</code></p>
 }}}