swr/rast: Add autogen of helper llvm intrinsics.

[mesa.git] / src / gallium / drivers / swr / rasterizer / jitter / fetch_jit.cpp

diff --git a/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp b/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp
index 9bb62976f60ba6dd5d7b8f15c498a17f588bc87c..5c8d81332df118affb7551937e4745172a92b9bc 100644 (file)
--- a/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp
+++ b/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp
@@ -253,7 +253,13 @@ Function* FetchJit::Create(const FETCH_COMPILE_STATE& fetchState)
                                                : vIndices2 = GetSimdValid32bitIndices(indices2, pLastIndex);
 #endif
             break; // incoming type is already 32bit int
-        default: SWR_INVALID("Unsupported index type"); vIndices = nullptr; break;
+        default:
+            SWR_INVALID("Unsupported index type");
+            vIndices = nullptr;
+#if USE_SIMD16_SHADERS
+            vIndices2 = nullptr;
+#endif
+            break;
     }
 
     if(fetchState.bForceSequentialAccessEnable)
@@ -434,6 +440,10 @@ void FetchJit::JitLoadVertices(const FETCH_COMPILE_STATE &fetchState, Value* str
         }
         else if (ied.InstanceStrideEnable)
         {
+            // silence unused variable warnings
+            startOffset = C(0);
+            vCurIndices = vIndices;
+
             SWR_ASSERT((0), "TODO: Fill out more once driver sends this down.");
         }
         else
@@ -445,7 +455,7 @@ void FetchJit::JitLoadVertices(const FETCH_COMPILE_STATE &fetchState, Value* str
         }
 
         // load SWR_VERTEX_BUFFER_STATE::pData
-        Value *stream = LOAD(streams, {ied.StreamIndex, SWR_VERTEX_BUFFER_STATE_pData});
+        Value *stream = LOAD(streams, {ied.StreamIndex, SWR_VERTEX_BUFFER_STATE_xpData});
 
         // load SWR_VERTEX_BUFFER_STATE::pitch
         Value *stride = LOAD(streams, {ied.StreamIndex, SWR_VERTEX_BUFFER_STATE_pitch});
@@ -933,10 +943,10 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
         SWR_ASSERT((info.bpp != 0), "Unsupported format in JitGatherVertices.");
         uint32_t bpc = info.bpp / info.numComps;  ///@todo Code below assumes all components are same size. Need to fix.
 
-        Value *stream = LOAD(streams, {ied.StreamIndex, SWR_VERTEX_BUFFER_STATE_pData});
+        Value *stream = LOAD(streams, {ied.StreamIndex, SWR_VERTEX_BUFFER_STATE_xpData});
 
         // VGATHER* takes an *i8 src pointer
-        Value *pStreamBase = BITCAST(stream, PointerType::get(mInt8Ty, 0));
+        Value *pStreamBase = INT_TO_PTR(stream, PointerType::get(mInt8Ty, 0));
 
         Value *stride = LOAD(streams, {ied.StreamIndex, SWR_VERTEX_BUFFER_STATE_pitch});
 #if USE_SIMD16_GATHERS
@@ -1039,6 +1049,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
         // calculate byte offset to the start of the VB
         Value* baseOffset = MUL(Z_EXT(startOffset, mInt64Ty), Z_EXT(stride, mInt64Ty));
         pStreamBase = GEP(pStreamBase, baseOffset);
+        Value* pStreamBaseGFX = ADD(stream, baseOffset);
 
         // if we have a start offset, subtract from max vertex. Used for OOB check
         maxVertex = SUB(Z_EXT(maxVertex, mInt64Ty), Z_EXT(startOffset, mInt64Ty));
@@ -1332,7 +1343,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
                                 // But, we know that elements must be aligned for FETCH. :)
                                 // Right shift the offset by a bit and then scale by 2 to remove the sign extension.
                                 Value *shiftedOffsets16 = LSHR(vOffsets16, 1);
-                                pVtxSrc2[currentVertexElement++] = GATHERPS_16(gatherSrc16, pStreamBase, shiftedOffsets16, vGatherMask16, 2);
+                                pVtxSrc2[currentVertexElement++] = GATHERPS_16(gatherSrc16, pStreamBaseGFX, shiftedOffsets16, vGatherMask16, 2, GFX_MEM_CLIENT_FETCH);
                             }
                             else
                             {
@@ -1349,9 +1360,6 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
                                 currentVertexElement = 0;
                             }
                         }
-
-                        // offset base to the next component in the vertex to gather
-                        pStreamBase = GEP(pStreamBase, C((char)4));
 #else
                         if (isComponentEnabled(compMask, i))
                         {
@@ -1364,7 +1372,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
                                 // But, we know that elements must be aligned for FETCH. :)
                                 // Right shift the offset by a bit and then scale by 2 to remove the sign extension.
                                 Value *vShiftedOffsets = LSHR(vOffsets, 1);
-                                vVertexElements[currentVertexElement++] = GATHERPS(gatherSrc, pStreamBase, vShiftedOffsets, vGatherMask, 2);
+                                vVertexElements[currentVertexElement++] = GATHERPS(gatherSrc, pStreamBaseGFX, vShiftedOffsets, vGatherMask, 2, GFX_MEM_CLIENT_FETCH);
                             }
                             else
                             {
@@ -1382,10 +1390,11 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
                                 currentVertexElement = 0;
                             }
                         }
+#endif
 
                         // offset base to the next component in the vertex to gather
                         pStreamBase = GEP(pStreamBase, C((char)4));
-#endif
+                        pStreamBaseGFX = ADD(pStreamBaseGFX, C((int64_t)4));
                     }
                 }
                     break;
@@ -1839,7 +1848,7 @@ Value* FetchJit::GetSimdValid16bitIndices(Value* pIndices, Value* pLastIndex)
 
         // if valid, load the index. if not, load 0 from the stack
         Value* pValid = SELECT(mask, pIndex, pZeroIndex);
-        Value *index = LOAD(pValid, "valid index");
+        Value *index = LOAD(pValid, "valid index", GFX_MEM_CLIENT_FETCH);
 
         // zero extended index to 32 bits and insert into the correct simd lane
         index = Z_EXT(index, mInt32Ty);
@@ -1875,13 +1884,11 @@ Value* FetchJit::GetSimdValid32bitIndices(Value* pIndices, Value* pLastIndex)
     //     vIndexMask    -1-1-1-1 0 0 0 0 : offsets < max pass
     //     vLoadedIndices 0 1 2 3 0 0 0 0 : offsets >= max masked to 0
     Value* vMaxIndex = VBROADCAST(numIndicesLeft);
-    Value* vIndexMask = VPCMPGTD(vMaxIndex,vIndexOffsets);
-
-    // VMASKLOAD takes an *i8 src pointer
-    pIndices = BITCAST(pIndices,PointerType::get(mInt8Ty,0));
+    Value* vIndexMask = ICMP_SGT(vMaxIndex, vIndexOffsets);
 
     // Load the indices; OOB loads 0
-    return MASKLOADD(pIndices,vIndexMask);
+    pIndices = BITCAST(pIndices, PointerType::get(mSimdInt32Ty, 0));
+    return MASKED_LOAD(pIndices, 4, vIndexMask, VIMMED1(0));
 }
 
 //////////////////////////////////////////////////////////////////////////