From a669c5ab36accbf73edf8b301728bab75f1dfa3d Mon Sep 17 00:00:00 2001
From: foreman <dl.constructicon@amd.com>
Date: Mon, 18 Aug 2014 16:46:45 -0400
Subject: [PATCH] P4 to Git Change 1067573 by skudchad@skudchad_opencl_win_2 on
 2014/08/18 16:38:03

	ECR #304775 - Refactor code to do line by line copies for read\write Rect. This avoids taking the blit copy path which may be even slower.

	ReviewBoardURL = http://ocltc.amd.com/reviews/r/5567/

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpublit.cpp#108 edit


[ROCm/clr commit: a5e788c9f8d0dae140a00ad2de755afb40801d41]
---
 .../clr/rocclr/runtime/device/gpu/gpublit.cpp | 67 ++++++++++---------
 1 file changed, 35 insertions(+), 32 deletions(-)

diff --git a/projects/clr/rocclr/runtime/device/gpu/gpublit.cpp b/projects/clr/rocclr/runtime/device/gpu/gpublit.cpp
index 9a64e6efed..4d638e5e31 100644
--- a/projects/clr/rocclr/runtime/device/gpu/gpublit.cpp
+++ b/projects/clr/rocclr/runtime/device/gpu/gpublit.cpp
@@ -613,7 +613,41 @@ DmaBlitManager::copyBufferRect(
         size_t  srcOffset;
         size_t  dstOffset;
 
-        if (!dev().settings().rectLinearDMA_) {
+        bool subWindowRectCopy = dev().settings().rectLinearDMA_;
+        // 19 bit limit in HW in SI and 16 bit limit in CI+(we adjust the ElementSize to 4bytes but the packet still has 14bits)
+        size_t pitchLimit = dev().settings().ciPlus_ ? 0xFFFF : 0x7FFFF;
+
+        srcOffset   = srcRect.offset(0, 0, 0);
+        dstOffset   = dstRect.offset(0, 0, 0);
+
+        if (subWindowRectCopy &&
+            (((srcOffset % 4) != 0) ||
+            ((dstOffset % 4) != 0) ||
+            ((size[0] % 4) != 0) ||
+            ((srcRect.rowPitch_ % 4) != 0) ||
+            ((srcRect.slicePitch_ % 4) != 0) ||
+            ((dstRect.rowPitch_ % 4) != 0) ||
+            ((dstRect.slicePitch_ % 4) != 0) ||
+            (srcRect.rowPitch_ > pitchLimit) ||
+            (dstRect.rowPitch_ > pitchLimit) ||
+            (size[0] > 0x3fff) ||   // 14 bits limit in HW
+            (size[1] > 0x3fff) ||   // 14 bits limit in HW
+            (size[2] > 0x7ff))) {    // 11 bits limit in HW
+            // Restriction with rectLinearDRMDMA packet
+            subWindowRectCopy = false;
+        }
+
+        if (subWindowRectCopy) {
+            // Copy data with subwindow copy packet
+            if (!gpuMem(srcMemory).partialMemCopyTo(gpu(),
+                amd::Coord3D(srcOffset, srcRect.rowPitch_, srcRect.slicePitch_),
+                amd::Coord3D(dstOffset, dstRect.rowPitch_, dstRect.slicePitch_),
+                size, gpuMem(dstMemory), true)) {
+                LogError("copyBufferRect failed!");
+                return false;
+            }
+        }
+        else {
             for (size_t z = 0; z < size[2]; ++z) {
                 for (size_t y = 0; y < size[1]; ++y) {
                     srcOffset   = srcRect.offset(0, y, z);
@@ -632,37 +666,6 @@ DmaBlitManager::copyBufferRect(
                 }
             }
         }
-        else {
-            srcOffset   = srcRect.offset(0, 0, 0);
-            dstOffset   = dstRect.offset(0, 0, 0);
-
-            // 19 bit limit in HW in SI and 16 bit limit in CI+(we adjust the ElementSize to 4bytes but the packet still has 14bits)
-            size_t pitchLimit = dev().settings().ciPlus_ ? 0xFFFF : 0x7FFFF;
-
-            if (((srcOffset % 4) != 0) ||
-                ((dstOffset % 4) != 0) ||
-                ((size[0] % 4) != 0) ||
-                ((srcRect.rowPitch_ % 4) != 0) ||
-                ((srcRect.slicePitch_ % 4) != 0) ||
-                ((dstRect.rowPitch_ % 4) != 0) ||
-                ((dstRect.slicePitch_ % 4) != 0) ||
-                (srcRect.rowPitch_ > pitchLimit) ||
-                (dstRect.rowPitch_ > pitchLimit) ||
-                (size[0] > 0x3fff) ||   // 14 bits limit in HW
-                (size[1] > 0x3fff) ||   // 14 bits limit in HW
-                (size[2] > 0x7ff)) {    // 11 bits limit in HW
-                // Restriction with rectLinearDRMDMA packet
-                return false;
-            }
-            // Copy data
-            if (!gpuMem(srcMemory).partialMemCopyTo(gpu(),
-                amd::Coord3D(srcOffset, srcRect.rowPitch_, srcRect.slicePitch_),
-                amd::Coord3D(dstOffset, dstRect.rowPitch_, dstRect.slicePitch_),
-                size, gpuMem(dstMemory), true)) {
-                LogError("copyBufferRect failed!");
-                return false;
-            }
-        }
     }
     return true;
 }