From 5a2bca5db5e025f0884487f590feac0c33db48fd Mon Sep 17 00:00:00 2001
From: Tim Rowley <timothy.o.rowley@intel.com>
Date: Wed, 20 Sep 2017 11:50:32 -0500
Subject: [PATCH] swr/rast: Handle instanceID offset / Instance Stride enable

Supported in JitGatherVertices(); FetchJit::JitLoadVertices() may require
similar changes, will need address this if it is determined that this
path is still in use.

Handle Force Sequential Access in FetchJit::Create.

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>
---
 .../swr/rasterizer/jitter/fetch_jit.cpp       | 46 ++++++++++++++++---
 1 file changed, 39 insertions(+), 7 deletions(-)

diff --git a/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp b/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp
index 906129829c8..1e3db902bb6 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp
+++ b/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp
@@ -222,6 +222,18 @@ Function* FetchJit::Create(const FETCH_COMPILE_STATE& fetchState)
         default: SWR_INVALID("Unsupported index type"); vIndices = nullptr; break;
     }
 
+    if(fetchState.bForceSequentialAccessEnable)
+    {
+        Value* pOffsets = C({ 0, 1, 2, 3, 4, 5, 6, 7 });
+
+        // VertexData buffers are accessed sequentially, the index is equal to the vertex number
+        vIndices = VBROADCAST(LOAD(mpFetchInfo, { 0, SWR_FETCH_CONTEXT_StartVertex }));
+        vIndices = ADD(vIndices, pOffsets);
+#if USE_SIMD16_SHADERS
+        vIndices2 = ADD(vIndices, VIMMED1(8));
+#endif
+    }
+
     Value* vVertexId = vIndices;
 #if USE_SIMD16_SHADERS
     Value* vVertexId2 = vIndices2;
@@ -275,12 +287,6 @@ Function* FetchJit::Create(const FETCH_COMPILE_STATE& fetchState)
                                  : JitGatherVertices(fetchState, streams, vIndices, pVtxOut);
 #endif
 
-    if (fetchState.bInstanceIDOffsetEnable)
-    {
-        // TODO: 
-        SWR_ASSERT((0), "Add support for handling InstanceID Offset Enable.");
-    }
-
     RET_VOID();
 
     JitManager::DumpToFile(fetch, "src");
@@ -362,6 +368,11 @@ void FetchJit::JitLoadVertices(const FETCH_COMPILE_STATE &fetchState, Value* str
 
         vectors.clear();
 
+        if (fetchState.bInstanceIDOffsetEnable)
+        {
+            SWR_ASSERT((0), "TODO: Fill out more once driver sends this down");
+        }
+
         Value *vCurIndices;
         Value *startOffset;
         if(ied.InstanceEnable)
@@ -831,8 +842,16 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
             minVertex = LOAD(minVertex);
         }
 
+        if (fetchState.bInstanceIDOffsetEnable)
+        {
+            // the InstanceID (curInstance) value is offset by StartInstanceLocation
+            curInstance = ADD(curInstance, startInstance);
+        }
+
         Value *vCurIndices;
         Value *startOffset;
+        Value *vInstanceStride = VIMMED1(0);
+
         if(ied.InstanceEnable)
         {
             Value* stepRate = C(ied.InstanceAdvancementState);
@@ -853,11 +872,19 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
         }
         else if (ied.InstanceStrideEnable)
         {
+            // grab the instance advancement state, determines stride in bytes from one instance to the next
+            Value* stepRate = C(ied.InstanceAdvancementState);
+            vInstanceStride = VBROADCAST(MUL(curInstance, stepRate));
+
+            // offset indices by baseVertex
+            vCurIndices = ADD(vIndices, vBaseVertex);
+
+            startOffset = startVertex;
             SWR_ASSERT((0), "TODO: Fill out more once driver sends this down.");
         }
         else
         {
-            // offset indices by baseVertex            
+            // offset indices by baseVertex
             vCurIndices = ADD(vIndices, vBaseVertex);
 
             startOffset = startVertex;
@@ -925,6 +952,11 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
         Value* vOffsets = MUL(vCurIndices, vStride);
         vOffsets = ADD(vOffsets, vAlignmentOffsets);
 
+        // if instance stride enable is:
+        //  true  - add product of the instanceID and advancement state to the offst into the VB
+        //  false - value of vInstanceStride has been initialialized to zero
+        vOffsets = ADD(vOffsets, vInstanceStride);
+
         // Packing and component control 
         ComponentEnable compMask = (ComponentEnable)ied.ComponentPacking;
         const ComponentControl compCtrl[4] { (ComponentControl)ied.ComponentControl0, (ComponentControl)ied.ComponentControl1, 
-- 
2.30.2