swr/rast: SIMD16 builder - cleanup naming (simd2 -> simd16)

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>

diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder.cpp b/src/gallium/drivers/swr/rasterizer/jitter/builder.cpp
index 4b83a3204cf7c3e4d2223b9368af26eedfedc1d0..c46159a35ac80811056e469fe11a612f7158fdf0 100644 (file)
--- a/src/gallium/drivers/swr/rasterizer/jitter/builder.cpp
+++ b/src/gallium/drivers/swr/rasterizer/jitter/builder.cpp
@@ -40,52 +40,56 @@ namespace SwrJit
     Builder::Builder(JitManager *pJitMgr)
         : mpJitMgr(pJitMgr)
     {
+        SWR_ASSERT(pJitMgr->mVWidth == 8);
+
         mVWidth = pJitMgr->mVWidth;
-#if USE_SIMD16_BUILDER
-        mVWidth2 = pJitMgr->mVWidth * 2;
-#endif
+        mVWidth16 = pJitMgr->mVWidth * 2;
 
         mpIRBuilder = &pJitMgr->mBuilder;
 
-        mVoidTy = Type::getVoidTy(pJitMgr->mContext);
-        mFP16Ty = Type::getHalfTy(pJitMgr->mContext);
-        mFP32Ty = Type::getFloatTy(pJitMgr->mContext);
-        mFP32PtrTy = PointerType::get(mFP32Ty, 0);
-        mDoubleTy = Type::getDoubleTy(pJitMgr->mContext);
-        mInt1Ty = Type::getInt1Ty(pJitMgr->mContext);
-        mInt8Ty = Type::getInt8Ty(pJitMgr->mContext);
-        mInt16Ty = Type::getInt16Ty(pJitMgr->mContext);
-        mInt32Ty = Type::getInt32Ty(pJitMgr->mContext);
-        mInt8PtrTy = PointerType::get(mInt8Ty, 0);
+        // Built in types: scalar
+
+        mVoidTy     = Type::getVoidTy(pJitMgr->mContext);
+        mFP16Ty     = Type::getHalfTy(pJitMgr->mContext);
+        mFP32Ty     = Type::getFloatTy(pJitMgr->mContext);
+        mFP32PtrTy  = PointerType::get(mFP32Ty, 0);
+        mDoubleTy   = Type::getDoubleTy(pJitMgr->mContext);
+        mInt1Ty     = Type::getInt1Ty(pJitMgr->mContext);
+        mInt8Ty     = Type::getInt8Ty(pJitMgr->mContext);
+        mInt16Ty    = Type::getInt16Ty(pJitMgr->mContext);
+        mInt32Ty    = Type::getInt32Ty(pJitMgr->mContext);
+        mInt8PtrTy  = PointerType::get(mInt8Ty, 0);
         mInt16PtrTy = PointerType::get(mInt16Ty, 0);
         mInt32PtrTy = PointerType::get(mInt32Ty, 0);
-        mInt64Ty = Type::getInt64Ty(pJitMgr->mContext);
-        mSimdInt1Ty = VectorType::get(mInt1Ty, mVWidth);
-        mSimdInt16Ty = VectorType::get(mInt16Ty, mVWidth);
-        mSimdInt32Ty = VectorType::get(mInt32Ty, mVWidth);
-        mSimdInt64Ty = VectorType::get(mInt64Ty, mVWidth);
-        mSimdFP16Ty = VectorType::get(mFP16Ty, mVWidth);
-        mSimdFP32Ty = VectorType::get(mFP32Ty, mVWidth);
-        mSimdVectorTy = ArrayType::get(mSimdFP32Ty, 4);
+        mInt64Ty    = Type::getInt64Ty(pJitMgr->mContext);
+
+        // Built in types: simd8
+
+        mSimdInt1Ty     = VectorType::get(mInt1Ty,  mVWidth);
+        mSimdInt16Ty    = VectorType::get(mInt16Ty, mVWidth);
+        mSimdInt32Ty    = VectorType::get(mInt32Ty, mVWidth);
+        mSimdInt64Ty    = VectorType::get(mInt64Ty, mVWidth);
+        mSimdFP16Ty     = VectorType::get(mFP16Ty,  mVWidth);
+        mSimdFP32Ty     = VectorType::get(mFP32Ty,  mVWidth);
+        mSimdVectorTy   = ArrayType::get(mSimdFP32Ty, 4);
         mSimdVectorTRTy = ArrayType::get(mSimdFP32Ty, 5);
-#if USE_SIMD16_BUILDER
-        mSimd2Int1Ty = VectorType::get(mInt1Ty, mVWidth2);
-        mSimd2Int16Ty = VectorType::get(mInt16Ty, mVWidth2);
-        mSimd2Int32Ty = VectorType::get(mInt32Ty, mVWidth2);
-        mSimd2Int64Ty = VectorType::get(mInt64Ty, mVWidth2);
-        mSimd2FP16Ty = VectorType::get(mFP16Ty, mVWidth2);
-        mSimd2FP32Ty = VectorType::get(mFP32Ty, mVWidth2);
-        mSimd2VectorTy = ArrayType::get(mSimd2FP32Ty, 4);
-        mSimd2VectorTRTy = ArrayType::get(mSimd2FP32Ty, 5);
-#endif
+
+        // Built in types: simd16
+
+        mSimd16Int1Ty       = VectorType::get(mInt1Ty,  mVWidth16);
+        mSimd16Int16Ty      = VectorType::get(mInt16Ty, mVWidth16);
+        mSimd16Int32Ty      = VectorType::get(mInt32Ty, mVWidth16);
+        mSimd16Int64Ty      = VectorType::get(mInt64Ty, mVWidth16);
+        mSimd16FP16Ty       = VectorType::get(mFP16Ty,  mVWidth16);
+        mSimd16FP32Ty       = VectorType::get(mFP32Ty,  mVWidth16);
+        mSimd16VectorTy     = ArrayType::get(mSimd16FP32Ty, 4);
+        mSimd16VectorTRTy   = ArrayType::get(mSimd16FP32Ty, 5);
 
         if (sizeof(uint32_t*) == 4)
         {
             mIntPtrTy = mInt32Ty;
             mSimdIntPtrTy = mSimdInt32Ty;
-#if USE_SIMD16_BUILDER
-            mSimd2IntPtrTy = mSimd2Int32Ty;
-#endif
+            mSimd16IntPtrTy = mSimd16Int32Ty;
         }
         else
         {
@@ -93,9 +97,7 @@ namespace SwrJit
 
             mIntPtrTy = mInt64Ty;
             mSimdIntPtrTy = mSimdInt64Ty;
-#if USE_SIMD16_BUILDER
-            mSimd2IntPtrTy = mSimd2Int64Ty;
-#endif
+            mSimd16IntPtrTy = mSimd16Int64Ty;
         }
     }
 }

diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder.h b/src/gallium/drivers/swr/rasterizer/jitter/builder.h
index c6ab64e06e8a2e9d67211fe8db35594af154a7d2..288c5d9ecd421f21ac093f592e157c98457508f7 100644 (file)
--- a/src/gallium/drivers/swr/rasterizer/jitter/builder.h
+++ b/src/gallium/drivers/swr/rasterizer/jitter/builder.h
@@ -32,26 +32,23 @@
 #include "JitManager.h"
 #include "common/formats.h"
 
-#define USE_SIMD16_BUILDER 0
-
 namespace SwrJit
 {
     using namespace llvm;
     struct Builder
     {
         Builder(JitManager *pJitMgr);
-        IRBuilder<>* IRB() { return mpIRBuilder; };
-        JitManager* JM() { return mpJitMgr; }
+        IRBuilder<> *IRB() { return mpIRBuilder; };
+        JitManager *JM() { return mpJitMgr; }
+
+        JitManager *mpJitMgr;
+        IRBuilder<> *mpIRBuilder;
 
-        JitManager* mpJitMgr;
-        IRBuilder<>* mpIRBuilder;
+        uint32_t             mVWidth;   // vector width simd8
+        uint32_t             mVWidth16; // vector width simd16
 
-        uint32_t             mVWidth;
-#if USE_SIMD16_BUILDER
-        uint32_t             mVWidth2;
-#endif
+        // Built in types: scalar
 
-        // Built in types.
         Type*                mVoidTy;
         Type*                mInt1Ty;
         Type*                mInt8Ty;
@@ -66,6 +63,9 @@ namespace SwrJit
         Type*                mInt8PtrTy;
         Type*                mInt16PtrTy;
         Type*                mInt32PtrTy;
+
+        // Built in types: simd8
+
         Type*                mSimdFP16Ty;
         Type*                mSimdFP32Ty;
         Type*                mSimdInt1Ty;
@@ -75,17 +75,18 @@ namespace SwrJit
         Type*                mSimdIntPtrTy;
         Type*                mSimdVectorTy;
         Type*                mSimdVectorTRTy;
-#if USE_SIMD16_BUILDER
-        Type*                mSimd2FP16Ty;
-        Type*                mSimd2FP32Ty;
-        Type*                mSimd2Int1Ty;
-        Type*                mSimd2Int16Ty;
-        Type*                mSimd2Int32Ty;
-        Type*                mSimd2Int64Ty;
-        Type*                mSimd2IntPtrTy;
-        Type*                mSimd2VectorTy;
-        Type*                mSimd2VectorTRTy;
-#endif
+
+        // Built in types: simd16
+
+        Type*                mSimd16FP16Ty;
+        Type*                mSimd16FP32Ty;
+        Type*                mSimd16Int1Ty;
+        Type*                mSimd16Int16Ty;
+        Type*                mSimd16Int32Ty;
+        Type*                mSimd16Int64Ty;
+        Type*                mSimd16IntPtrTy;
+        Type*                mSimd16VectorTy;
+        Type*                mSimd16VectorTRTy;
 
 #include "gen_builder.hpp"
 #include "gen_builder_x86.hpp"

diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp b/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp
index 0774889af105b4a0dbaa13e7c32705aea47c63de..92b07a5bfc05d3995ef156221c77dcc7ec2e0647 100644 (file)
--- a/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp
+++ b/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp
@@ -196,56 +196,59 @@ namespace SwrJit
         return ConstantVector::getSplat(mVWidth, cast<ConstantInt>(C(i)));
     }
 
-    Value *Builder::VIMMED1(uint32_t i)
+    Value *Builder::VIMMED1_16(int i)
     {
-        return ConstantVector::getSplat(mVWidth, cast<ConstantInt>(C(i)));
+        return ConstantVector::getSplat(mVWidth16, cast<ConstantInt>(C(i)));
     }
 
-    Value *Builder::VIMMED1(float i)
+    Value *Builder::VIMMED1(uint32_t i)
     {
-        return ConstantVector::getSplat(mVWidth, cast<ConstantFP>(C(i)));
+        return ConstantVector::getSplat(mVWidth, cast<ConstantInt>(C(i)));
     }
 
-    Value *Builder::VIMMED1(bool i)
+    Value *Builder::VIMMED1_16(uint32_t i)
     {
-        return ConstantVector::getSplat(mVWidth, cast<ConstantInt>(C(i)));
+        return ConstantVector::getSplat(mVWidth16, cast<ConstantInt>(C(i)));
     }
 
-#if USE_SIMD16_BUILDER
-    Value *Builder::VIMMED2_1(int i)
+    Value *Builder::VIMMED1(float i)
     {
-        return ConstantVector::getSplat(mVWidth2, cast<ConstantInt>(C(i)));
+        return ConstantVector::getSplat(mVWidth, cast<ConstantFP>(C(i)));
     }
 
-    Value *Builder::VIMMED2_1(uint32_t i)
+    Value *Builder::VIMMED1_16(float i)
     {
-        return ConstantVector::getSplat(mVWidth2, cast<ConstantInt>(C(i)));
+        return ConstantVector::getSplat(mVWidth16, cast<ConstantFP>(C(i)));
     }
 
-    Value *Builder::VIMMED2_1(float i)
+    Value *Builder::VIMMED1(bool i)
     {
-        return ConstantVector::getSplat(mVWidth2, cast<ConstantFP>(C(i)));
+        return ConstantVector::getSplat(mVWidth, cast<ConstantInt>(C(i)));
     }
 
-    Value *Builder::VIMMED2_1(bool i)
+    Value *Builder::VIMMED1_16(bool i)
     {
-        return ConstantVector::getSplat(mVWidth2, cast<ConstantInt>(C(i)));
+        return ConstantVector::getSplat(mVWidth16, cast<ConstantInt>(C(i)));
     }
 
-#endif
     Value *Builder::VUNDEF_IPTR()
     {
         return UndefValue::get(VectorType::get(mInt32PtrTy,mVWidth));
     }
 
+    Value *Builder::VUNDEF(Type* t)
+    {
+        return UndefValue::get(VectorType::get(t, mVWidth));
+    }
+
     Value *Builder::VUNDEF_I()
     {
         return UndefValue::get(VectorType::get(mInt32Ty, mVWidth));
     }
 
-    Value *Builder::VUNDEF(Type *ty, uint32_t size)
+    Value *Builder::VUNDEF_I_16()
     {
-        return UndefValue::get(VectorType::get(ty, size));
+        return UndefValue::get(VectorType::get(mInt32Ty, mVWidth16));
     }
 
     Value *Builder::VUNDEF_F()
@@ -253,21 +256,14 @@ namespace SwrJit
         return UndefValue::get(VectorType::get(mFP32Ty, mVWidth));
     }
 
-#if USE_SIMD16_BUILDER
-    Value *Builder::VUNDEF2_F()
-    {
-        return UndefValue::get(VectorType::get(mFP32Ty, mVWidth2));
-    }
-
-    Value *Builder::VUNDEF2_I()
+    Value *Builder::VUNDEF_F_16()
     {
-        return UndefValue::get(VectorType::get(mInt32Ty, mVWidth2));
+        return UndefValue::get(VectorType::get(mFP32Ty, mVWidth16));
     }
 
-#endif
-    Value *Builder::VUNDEF(Type* t)
+    Value *Builder::VUNDEF(Type *ty, uint32_t size)
     {
-        return UndefValue::get(VectorType::get(t, mVWidth));
+        return UndefValue::get(VectorType::get(ty, size));
     }
 
     Value *Builder::VBROADCAST(Value *src)
@@ -281,8 +277,7 @@ namespace SwrJit
         return VECTOR_SPLAT(mVWidth, src);
     }
 
-#if USE_SIMD16_BUILDER
-    Value *Builder::VBROADCAST2(Value *src)
+    Value *Builder::VBROADCAST_16(Value *src)
     {
         // check if src is already a vector
         if (src->getType()->isVectorTy())
@@ -290,10 +285,9 @@ namespace SwrJit
             return src;
         }
 
-        return VECTOR_SPLAT(mVWidth2, src);
+        return VECTOR_SPLAT(mVWidth16, src);
     }
 
-#endif
     uint32_t Builder::IMMED(Value* v)
     {
         SWR_ASSERT(isa<ConstantInt>(v));
@@ -632,18 +626,18 @@ namespace SwrJit
                 Value *val = LOAD(validAddress);
                 vGather = VINSERT(vGather,val,C(i));
             }
+
             STACKRESTORE(pStack);
         }
 
         return vGather;
     }
 
-#if USE_SIMD16_BUILDER
     Value *Builder::GATHERPS_16(Value *vSrc, Value *pBase, Value *vIndices, Value *vMask, uint8_t scale)
     {
-        Value *vGather = VUNDEF2_F();
+        Value *vGather = VUNDEF_F_16();
 
-        // use avx512 gather instruction if available
+        // use AVX512F gather instruction if available
         if (JM()->mArch.AVX512F())
         {
             // force mask to <N-bit Integer>, required by vgather2
@@ -653,25 +647,24 @@ namespace SwrJit
         }
         else
         {
-            Value *src0 = EXTRACT2(vSrc, 0);
-            Value *src1 = EXTRACT2(vSrc, 1);
+            Value *src0 = EXTRACT_16(vSrc, 0);
+            Value *src1 = EXTRACT_16(vSrc, 1);
 
-            Value *indices0 = EXTRACT2(vIndices, 0);
-            Value *indices1 = EXTRACT2(vIndices, 1);
+            Value *indices0 = EXTRACT_16(vIndices, 0);
+            Value *indices1 = EXTRACT_16(vIndices, 1);
 
-            Value *mask0 = EXTRACT2(vMask, 0);
-            Value *mask1 = EXTRACT2(vMask, 1);
+            Value *mask0 = EXTRACT_16(vMask, 0);
+            Value *mask1 = EXTRACT_16(vMask, 1);
 
             Value *gather0 = GATHERPS(src0, pBase, indices0, mask0, scale);
             Value *gather1 = GATHERPS(src1, pBase, indices1, mask1, scale);
 
-            vGather = JOIN2(gather0, gather1);
+            vGather = JOIN_16(gather0, gather1);
         }
 
         return vGather;
     }
 
-#endif
     //////////////////////////////////////////////////////////////////////////
     /// @brief Generate a masked gather operation in LLVM IR.  If not  
     /// supported on the underlying platform, emulate it with loads
@@ -718,15 +711,15 @@ namespace SwrJit
 
             STACKRESTORE(pStack);
         }
+
         return vGather;
     }
 
-#if USE_SIMD16_BUILDER
     Value *Builder::GATHERDD_16(Value *vSrc, Value *pBase, Value *vIndices, Value *vMask, uint8_t scale)
     {
-        Value *vGather = VUNDEF2_F();
+        Value *vGather = VUNDEF_I_16();
 
-        // use avx512 gather instruction if available
+        // use AVX512F gather instruction if available
         if (JM()->mArch.AVX512F())
         {
             // force mask to <N-bit Integer>, required by vgather2
@@ -736,25 +729,24 @@ namespace SwrJit
         }
         else
         {
-            Value *src0 = EXTRACT2(vSrc, 0);
-            Value *src1 = EXTRACT2(vSrc, 1);
+            Value *src0 = EXTRACT_16(vSrc, 0);
+            Value *src1 = EXTRACT_16(vSrc, 1);
 
-            Value *indices0 = EXTRACT2(vIndices, 0);
-            Value *indices1 = EXTRACT2(vIndices, 1);
+            Value *indices0 = EXTRACT_16(vIndices, 0);
+            Value *indices1 = EXTRACT_16(vIndices, 1);
 
-            Value *mask0 = EXTRACT2(vMask, 0);
-            Value *mask1 = EXTRACT2(vMask, 1);
+            Value *mask0 = EXTRACT_16(vMask, 0);
+            Value *mask1 = EXTRACT_16(vMask, 1);
 
             Value *gather0 = GATHERDD(src0, pBase, indices0, mask0, scale);
             Value *gather1 = GATHERDD(src1, pBase, indices1, mask1, scale);
 
-            vGather = JOIN2(gather0, gather1);
+            vGather = JOIN_16(gather0, gather1);
         }
 
         return vGather;
     }
 
-#endif
     //////////////////////////////////////////////////////////////////////////
     /// @brief Generate a masked gather operation in LLVM IR.  If not
     /// supported on the underlying platform, emulate it with loads
@@ -804,21 +796,22 @@ namespace SwrJit
         return vGather;
     }
 
-#if USE_SIMD16_BUILDER
-    Value *Builder::EXTRACT2(Value *x, uint32_t imm)
+    Value *Builder::EXTRACT_16(Value *x, uint32_t imm)
     {
         if (imm == 0)
-            return VSHUFFLE(x, UndefValue::get(x->getType()), {0, 1, 2, 3, 4, 5, 6, 7});
+        {
+            return VSHUFFLE(x, UndefValue::get(x->getType()), { 0, 1, 2, 3, 4, 5, 6, 7 });
+        }
         else
-            return VSHUFFLE(x, UndefValue::get(x->getType()), {8, 9, 10, 11, 12, 13, 14, 15});
+        {
+            return VSHUFFLE(x, UndefValue::get(x->getType()), { 8, 9, 10, 11, 12, 13, 14, 15 });
+        }
     }
 
-    Value *Builder::JOIN2(Value *a, Value *b)
+    Value *Builder::JOIN_16(Value *a, Value *b)
     {
-        return VSHUFFLE(a, b,
-                        {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15});
+        return VSHUFFLE(a, b, { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 });
     }
-#endif
 
     //////////////////////////////////////////////////////////////////////////
     /// @brief convert x86 <N x float> mask to llvm <N x i1> mask
@@ -828,14 +821,12 @@ namespace SwrJit
         return ICMP_SLT(src, VIMMED1(0));
     }
 
-#if USE_SIMD16_BUILDER
-    Value *Builder::MASK2(Value *vmask)
+    Value *Builder::MASK_16(Value *vmask)
     {
-        Value *src = BITCAST(vmask, mSimd2Int32Ty);
-        return ICMP_SLT(src, VIMMED2_1(0));
+        Value *src = BITCAST(vmask, mSimd16Int32Ty);
+        return ICMP_SLT(src, VIMMED1_16(0));
     }
 
-#endif
     //////////////////////////////////////////////////////////////////////////
     /// @brief convert llvm <N x i1> mask to x86 <N x i32> mask
     Value *Builder::VMASK(Value *mask)
@@ -843,13 +834,11 @@ namespace SwrJit
         return S_EXT(mask, mSimdInt32Ty);
     }
 
-#if USE_SIMD16_BUILDER
-    Value *Builder::VMASK2(Value *mask)
+    Value *Builder::VMASK_16(Value *mask)
     {
-        return S_EXT(mask, mSimd2Int32Ty);
+        return S_EXT(mask, mSimd16Int32Ty);
     }
 
-#endif
     //////////////////////////////////////////////////////////////////////////
     /// @brief Generate a VPSHUFB operation in LLVM IR.  If not  
     /// supported on the underlying platform, emulate it

diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.h b/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.h
index 646ed0efb2e976f0770e7603edc86516e06dcd19..16ff693c6e4a7107856f4657b1c3690cf7cf16af 100644 (file)
--- a/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.h
+++ b/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.h
@@ -50,29 +50,34 @@ Constant *C(const std::initializer_list<Ty> &constList)
 }
 
 Constant *PRED(bool pred);
+
 Value *VIMMED1(int i);
+Value *VIMMED1_16(int i);
+
 Value *VIMMED1(uint32_t i);
+Value *VIMMED1_16(uint32_t i);
+
 Value *VIMMED1(float i);
+Value *VIMMED1_16(float i);
+
 Value *VIMMED1(bool i);
-#if USE_SIMD16_BUILDER
-Value *VIMMED2_1(int i);
-Value *VIMMED2_1(uint32_t i);
-Value *VIMMED2_1(float i);
-Value *VIMMED2_1(bool i);
-#endif
+Value *VIMMED1_16(bool i);
+
 Value *VUNDEF(Type* t);
+
 Value *VUNDEF_F();
+Value *VUNDEF_F_16();
+
 Value *VUNDEF_I();
-#if USE_SIMD16_BUILDER
-Value *VUNDEF2_F();
-Value *VUNDEF2_I();
-#endif
+Value *VUNDEF_I_16();
+
 Value *VUNDEF(Type* ty, uint32_t size);
+
 Value *VUNDEF_IPTR();
+
 Value *VBROADCAST(Value *src);
-#if USE_SIMD16_BUILDER
-Value *VBROADCAST2(Value *src);
-#endif
+Value *VBROADCAST_16(Value *src);
+
 Value *VRCP(Value *va);
 Value *VPLANEPS(Value* vA, Value* vB, Value* vC, Value* &vX, Value* &vY);
 
@@ -105,21 +110,18 @@ Value *VCMPPS_GT(Value* a, Value* b)    { return VCMPPS(a, b, C((uint8_t)_CMP_GT
 Value *VCMPPS_NOTNAN(Value* a, Value* b){ return VCMPPS(a, b, C((uint8_t)_CMP_ORD_Q)); }
 
 Value *MASK(Value *vmask);
+Value *MASK_16(Value *vmask);
+
 Value *VMASK(Value *mask);
-#if USE_SIMD16_BUILDER
-Value *MASK2(Value *vmask);
-Value *VMASK2(Value *mask);
-#endif
+Value *VMASK_16(Value *mask);
 
 //////////////////////////////////////////////////////////////////////////
 /// @brief functions that build IR to call x86 intrinsics directly, or
 /// emulate them with other instructions if not available on the host
 //////////////////////////////////////////////////////////////////////////
 
-#if USE_SIMD16_BUILDER
-Value *EXTRACT2(Value *x, uint32_t imm);
-Value *JOIN2(Value *a, Value *b);
-#endif
+Value *EXTRACT_16(Value *x, uint32_t imm);
+Value *JOIN_16(Value *a, Value *b);
 
 Value *MASKLOADD(Value* src, Value* mask);
 
@@ -127,16 +129,14 @@ void Gather4(const SWR_FORMAT format, Value* pSrcBase, Value* byteOffsets,
                       Value* mask, Value* vGatherComponents[], bool bPackedOutput);
 
 Value *GATHERPS(Value *src, Value *pBase, Value *indices, Value *mask, uint8_t scale = 1);
-#if USE_SIMD16_BUILDER
 Value *GATHERPS_16(Value *src, Value *pBase, Value *indices, Value *mask, uint8_t scale = 1);
-#endif
+
 void GATHER4PS(const SWR_FORMAT_INFO &info, Value* pSrcBase, Value* byteOffsets,
                Value* mask, Value* vGatherComponents[], bool bPackedOutput);
 
 Value *GATHERDD(Value* src, Value* pBase, Value* indices, Value* mask, uint8_t scale = 1);
-#if USE_SIMD16_BUILDER
 Value *GATHERDD_16(Value *src, Value *pBase, Value *indices, Value *mask, uint8_t scale = 1);
-#endif
+
 void GATHER4DD(const SWR_FORMAT_INFO &info, Value* pSrcBase, Value* byteOffsets,
                Value* mask, Value* vGatherComponents[], bool bPackedOutput);
 

diff --git a/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp b/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp
index aa911b58f3fa6a1d8a1b0b7dde7674f7deed0cb8..c7605046c35ca9810c7a49a75693bb6835dad338 100644 (file)
--- a/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp
+++ b/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp
@@ -49,6 +49,11 @@ enum ConversionType
     CONVERT_SFIXED,
 };
 
+#if USE_SIMD16_SHADERS
+#define USE_SIMD16_GATHERS 0
+#define USE_SIMD16_BUILDER 0
+#endif
+
 //////////////////////////////////////////////////////////////////////////
 /// Interface to Jitting a fetch shader
 //////////////////////////////////////////////////////////////////////////
@@ -101,7 +106,6 @@ struct FetchJit : public Builder
 
     void JitLoadVertices(const FETCH_COMPILE_STATE &fetchState, Value* streams, Value* vIndices, Value* pVtxOut);
 #if USE_SIMD16_SHADERS
-#define USE_SIMD16_GATHERS 0
 
 #if USE_SIMD16_GATHERS
     void JitGatherVertices(const FETCH_COMPILE_STATE &fetchState, Value *streams, Value *vIndices, Value *vIndices2, Value *pVtxOut, bool useVertexID2);
@@ -150,7 +154,7 @@ Function* FetchJit::Create(const FETCH_COMPILE_STATE& fetchState)
     pVtxOut = GEP(pVtxOut, C(0));
 #if USE_SIMD16_SHADERS
 #if 0// USE_SIMD16_BUILDER
-    pVtxOut = BITCAST(pVtxOut, PointerType::get(VectorType::get(mFP32Ty, mVWidth2), 0));
+    pVtxOut = BITCAST(pVtxOut, PointerType::get(VectorType::get(mFP32Ty, mVWidth16), 0));
 #else
     pVtxOut = BITCAST(pVtxOut, PointerType::get(VectorType::get(mFP32Ty, mVWidth), 0));
 #endif
@@ -841,7 +845,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
     Value* curInstance = LOAD(mpFetchInfo, {0, SWR_FETCH_CONTEXT_CurInstance});
 #if USE_SIMD16_GATHERS
 #if USE_SIMD16_BUILDER
-    Value* vBaseVertex16 = VBROADCAST2(LOAD(mpFetchInfo, { 0, SWR_FETCH_CONTEXT_BaseVertex }));
+    Value* vBaseVertex16 = VBROADCAST_16(LOAD(mpFetchInfo, { 0, SWR_FETCH_CONTEXT_BaseVertex }));
 #else
     Value* vBaseVertex = VBROADCAST(LOAD(mpFetchInfo, { 0, SWR_FETCH_CONTEXT_BaseVertex }));
 #endif
@@ -872,7 +876,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
         Value *stride = LOAD(streams, {ied.StreamIndex, SWR_VERTEX_BUFFER_STATE_pitch});
 #if USE_SIMD16_GATHERS
 #if USE_SIMD16_BUILDER
-        Value *vStride16 = VBROADCAST2(stride);
+        Value *vStride16 = VBROADCAST_16(stride);
 #else
         Value *vStride = VBROADCAST(stride);
 #endif
@@ -908,7 +912,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
         Value *startOffset;
 #if USE_SIMD16_GATHERS
 #if USE_SIMD16_BUILDER
-        Value *vInstanceStride16 = VIMMED2_1(0);
+        Value *vInstanceStride16 = VIMMED1_16(0);
 #else
         Value *vInstanceStride = VIMMED1(0);
 #endif
@@ -932,7 +936,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
 
 #if USE_SIMD16_GATHERS
 #if USE_SIMD16_BUILDER
-            vCurIndices16 = VBROADCAST2(calcInstance);
+            vCurIndices16 = VBROADCAST_16(calcInstance);
 #else
             vCurIndices = VBROADCAST(calcInstance);
             vCurIndices2 = VBROADCAST(calcInstance);
@@ -949,7 +953,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
             Value* stepRate = C(ied.InstanceAdvancementState);
 #if USE_SIMD16_GATHERS
 #if USE_SIMD16_BUILDER
-            vInstanceStride16 = VBROADCAST2(MUL(curInstance, stepRate));
+            vInstanceStride16 = VBROADCAST_16(MUL(curInstance, stepRate));
 #else
             vInstanceStride = VBROADCAST(MUL(curInstance, stepRate));
 #endif
@@ -960,7 +964,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
             // offset indices by baseVertex
 #if USE_SIMD16_GATHERS
 #if USE_SIMD16_BUILDER
-            Value *vIndices16 = JOIN2(vIndices, vIndices2);
+            Value *vIndices16 = JOIN_16(vIndices, vIndices2);
 
             vCurIndices16 = ADD(vIndices16, vBaseVertex16);
 #else
@@ -979,7 +983,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
             // offset indices by baseVertex
 #if USE_SIMD16_GATHERS
 #if USE_SIMD16_BUILDER
-            Value *vIndices16 = JOIN2(vIndices, vIndices2);
+            Value *vIndices16 = JOIN_16(vIndices, vIndices2);
 
             vCurIndices16 = ADD(vIndices16, vBaseVertex16);
 #else
@@ -1019,9 +1023,9 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
         partialInboundsSize = LOAD(partialInboundsSize);
 #if USE_SIMD16_GATHERS
 #if USE_SIMD16_BUILDER
-        Value *vPartialVertexSize = VBROADCAST2(partialInboundsSize);
-        Value *vBpp = VBROADCAST2(C(info.Bpp));
-        Value *vAlignmentOffsets = VBROADCAST2(C(ied.AlignedByteOffset));
+        Value *vPartialVertexSize = VBROADCAST_16(partialInboundsSize);
+        Value *vBpp = VBROADCAST_16(C(info.Bpp));
+        Value *vAlignmentOffsets = VBROADCAST_16(C(ied.AlignedByteOffset));
 #else
         Value *vPartialVertexSize = VBROADCAST(partialInboundsSize);
         Value *vBpp = VBROADCAST(C(info.Bpp));
@@ -1039,11 +1043,11 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
 #if USE_SIMD16_GATHERS
 #if USE_SIMD16_BUILDER
         // override cur indices with 0 if pitch is 0
-        Value *pZeroPitchMask16 = ICMP_EQ(vStride16, VIMMED2_1(0));
-        vCurIndices16 = SELECT(pZeroPitchMask16, VIMMED2_1(0), vCurIndices16);
+        Value *pZeroPitchMask16 = ICMP_EQ(vStride16, VIMMED1_16(0));
+        vCurIndices16 = SELECT(pZeroPitchMask16, VIMMED1_16(0), vCurIndices16);
 
         // are vertices partially OOB?
-        Value *vMaxVertex16 = VBROADCAST2(maxVertex);
+        Value *vMaxVertex16 = VBROADCAST_16(maxVertex);
         Value *vPartialOOBMask = ICMP_EQ(vCurIndices16, vMaxVertex16);
 
         // are vertices fully in bounds?
@@ -1054,7 +1058,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
         if (fetchState.bPartialVertexBuffer)
         {
             // are vertices below minVertex limit?
-            Value *vMinVertex16 = VBROADCAST2(minVertex);
+            Value *vMinVertex16 = VBROADCAST_16(minVertex);
             Value *vMinGatherMask16 = ICMP_UGE(vCurIndices16, vMinVertex16);
 
             // only fetch lanes that pass both tests
@@ -1079,11 +1083,11 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
 
         // TODO: remove the following simd8 interop stuff once all code paths are fully widened to SIMD16..
 
-        Value *vGatherMask = EXTRACT2(vGatherMask16, 0);
-        Value *vGatherMask2 = EXTRACT2(vGatherMask16, 1);
+        Value *vGatherMask  = EXTRACT_16(vGatherMask16, 0);
+        Value *vGatherMask2 = EXTRACT_16(vGatherMask16, 1);
 
-        Value *vOffsets = EXTRACT2(vOffsets16, 0);
-        Value *vOffsets2 = EXTRACT2(vOffsets16, 1);
+        Value *vOffsets  = EXTRACT_16(vOffsets16, 0);
+        Value *vOffsets2 = EXTRACT_16(vOffsets16, 1);
 #else
         // override cur indices with 0 if pitch is 0
         Value* pZeroPitchMask = ICMP_EQ(vStride, VIMMED1(0));
@@ -1198,7 +1202,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
                 {
 #if USE_SIMD16_BUILDER
                     // pack adjacent pairs of SIMD8s into SIMD16s
-                    pVtxSrc2[currentVertexElement] = JOIN2(pResults[c], pResults2[c]);
+                    pVtxSrc2[currentVertexElement] = JOIN_16(pResults[c], pResults2[c]);
 
 #else
                     vVertexElements[currentVertexElement]  = pResults[c];
@@ -1211,7 +1215,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
                     {
 #if USE_SIMD16_BUILDER
                         // store SIMD16s
-                        Value *pVtxOut2 = BITCAST(pVtxOut, PointerType::get(VectorType::get(mFP32Ty, mVWidth2), 0));
+                        Value *pVtxOut2 = BITCAST(pVtxOut, PointerType::get(VectorType::get(mFP32Ty, mVWidth16), 0));
 
                         StoreVertexElements2(pVtxOut2, outputElt, 4, pVtxSrc2);
 
@@ -1254,7 +1258,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
 #if USE_SIMD16_GATHERS
             Value *gatherSrc2 = VIMMED1(0.0f);
 #if USE_SIMD16_BUILDER
-            Value *gatherSrc16 = VIMMED2_1(0.0f);
+            Value *gatherSrc16 = VIMMED1_16(0.0f);
 #endif
 #endif
 
@@ -1282,7 +1286,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
                     }
                     else
                     {
-                        gatherResult[0] = VUNDEF2_I();
+                        gatherResult[0] = VUNDEF_I_16();
                     }
 
                     // if we have at least one component out of z or w to fetch
@@ -1300,7 +1304,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
                     }
                     else
                     {
-                        gatherResult[1] = VUNDEF2_I();
+                        gatherResult[1] = VUNDEF_I_16();
                     }
 
 #else
@@ -1347,7 +1351,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
                     if (compMask)
                     {
 #if USE_SIMD16_BUILDER
-                        Value *pVtxOut2 = BITCAST(pVtxOut, PointerType::get(VectorType::get(mFP32Ty, mVWidth2), 0));
+                        Value *pVtxOut2 = BITCAST(pVtxOut, PointerType::get(VectorType::get(mFP32Ty, mVWidth16), 0));
 
                         Shuffle16bpcArgs args = std::forward_as_tuple(gatherResult, pVtxOut2, Instruction::CastOps::FPExt, CONVERT_NONE,
                             currentVertexElement, outputElt, compMask, compCtrl, pVtxSrc2);
@@ -1432,7 +1436,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
 
 #if USE_SIMD16_BUILDER
                                 // pack adjacent pairs of SIMD8s into SIMD16s
-                                pVtxSrc2[currentVertexElement] = JOIN2(vVertexElements[currentVertexElement],
+                                pVtxSrc2[currentVertexElement] = JOIN_16(vVertexElements[currentVertexElement],
                                                                        vVertexElements2[currentVertexElement]);
 
 #endif
@@ -1449,7 +1453,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
 
 #if USE_SIMD16_BUILDER
                                 // pack adjacent pairs of SIMD8s into SIMD16s
-                                pVtxSrc2[currentVertexElement] = JOIN2(vVertexElements[currentVertexElement],
+                                pVtxSrc2[currentVertexElement] = JOIN_16(vVertexElements[currentVertexElement],
                                                                        vVertexElements2[currentVertexElement]);
 
 #endif
@@ -1461,7 +1465,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
                             {
 #if USE_SIMD16_BUILDER
                                 // store SIMD16s
-                                Value *pVtxOut2 = BITCAST(pVtxOut, PointerType::get(VectorType::get(mFP32Ty, mVWidth2), 0));
+                                Value *pVtxOut2 = BITCAST(pVtxOut, PointerType::get(VectorType::get(mFP32Ty, mVWidth16), 0));
 
                                 StoreVertexElements2(pVtxOut2, outputElt, 4, pVtxSrc2);
 
@@ -1553,7 +1557,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
 
 #if USE_SIMD16_BUILDER
                                 // pack adjacent pairs of SIMD8s into SIMD16s
-                                pVtxSrc2[currentVertexElement] = JOIN2(pGather, pGather2);
+                                pVtxSrc2[currentVertexElement] = JOIN_16(pGather, pGather2);
 
 #else
                                 vVertexElements[currentVertexElement]  = pGather;
@@ -1579,7 +1583,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
                             {
 #if USE_SIMD16_BUILDER
                                 // store SIMD16s
-                                Value *pVtxOut2 = BITCAST(pVtxOut, PointerType::get(VectorType::get(mFP32Ty, mVWidth2), 0));
+                                Value *pVtxOut2 = BITCAST(pVtxOut, PointerType::get(VectorType::get(mFP32Ty, mVWidth16), 0));
 
                                 StoreVertexElements2(pVtxOut2, outputElt, 4, pVtxSrc2);
 
@@ -1692,7 +1696,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
 #if USE_SIMD16_GATHERS
             Value* gatherSrc2 = VIMMED1(0);
 #if USE_SIMD16_BUILDER
-            Value *gatherSrc16 = VIMMED2_1(0);
+            Value *gatherSrc16 = VIMMED1_16(0);
 #endif
 #endif
 
@@ -1722,7 +1726,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
 
 #endif
 #if USE_SIMD16_BUILDER
-                        Value *pVtxOut2 = BITCAST(pVtxOut, PointerType::get(VectorType::get(mFP32Ty, mVWidth2), 0));
+                        Value *pVtxOut2 = BITCAST(pVtxOut, PointerType::get(VectorType::get(mFP32Ty, mVWidth16), 0));
 
                         Shuffle8bpcArgs args = std::forward_as_tuple(gatherResult, pVtxOut2, extendCastType, conversionType,
                             currentVertexElement, outputElt, compMask, compCtrl, pVtxSrc2, info.swizzle);
@@ -1776,7 +1780,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
                     }
                     else
                     {
-                        gatherResult[0] = VUNDEF2_I();
+                        gatherResult[0] = VUNDEF_I_16();
                     }
 
                     // if we have at least one component out of z or w to fetch
@@ -1794,7 +1798,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
                     }
                     else
                     {
-                        gatherResult[1] = VUNDEF2_I();
+                        gatherResult[1] = VUNDEF_I_16();
                     }
 
 #else
@@ -1841,7 +1845,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
                     if (compMask)
                     {
 #if USE_SIMD16_BUILDER
-                        Value *pVtxOut2 = BITCAST(pVtxOut, PointerType::get(VectorType::get(mFP32Ty, mVWidth2), 0));
+                        Value *pVtxOut2 = BITCAST(pVtxOut, PointerType::get(VectorType::get(mFP32Ty, mVWidth16), 0));
 
                         Shuffle16bpcArgs args = std::forward_as_tuple(gatherResult, pVtxOut2, extendCastType, conversionType,
                             currentVertexElement, outputElt, compMask, compCtrl, pVtxSrc2);
@@ -1914,15 +1918,15 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
 
                                 if (conversionType == CONVERT_USCALED)
                                 {
-                                    pGather = UI_TO_FP(pGather, mSimd2FP32Ty);
+                                    pGather = UI_TO_FP(pGather, mSimd16FP32Ty);
                                 }
                                 else if (conversionType == CONVERT_SSCALED)
                                 {
-                                    pGather = SI_TO_FP(pGather, mSimd2FP32Ty);
+                                    pGather = SI_TO_FP(pGather, mSimd16FP32Ty);
                                 }
                                 else if (conversionType == CONVERT_SFIXED)
                                 {
-                                    pGather = FMUL(SI_TO_FP(pGather, mSimd2FP32Ty), VBROADCAST2(C(1 / 65536.0f)));
+                                    pGather = FMUL(SI_TO_FP(pGather, mSimd16FP32Ty), VBROADCAST_16(C(1 / 65536.0f)));
                                 }
 
 #else
@@ -2008,7 +2012,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
 #if USE_SIMD16_GATHERS
 #if USE_SIMD16_BUILDER
                                 // store SIMD16s
-                                Value *pVtxOut2 = BITCAST(pVtxOut, PointerType::get(VectorType::get(mFP32Ty, mVWidth2), 0));
+                                Value *pVtxOut2 = BITCAST(pVtxOut, PointerType::get(VectorType::get(mFP32Ty, mVWidth16), 0));
 
                                 StoreVertexElements2(pVtxOut2, outputElt, 4, pVtxSrc2);
 
@@ -2043,7 +2047,7 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
 #if USE_SIMD16_GATHERS
 #if USE_SIMD16_BUILDER
         // store SIMD16s
-        Value *pVtxOut2 = BITCAST(pVtxOut, PointerType::get(VectorType::get(mFP32Ty, mVWidth2), 0));
+        Value *pVtxOut2 = BITCAST(pVtxOut, PointerType::get(VectorType::get(mFP32Ty, mVWidth16), 0));
 
         StoreVertexElements2(pVtxOut2, outputElt, currentVertexElement, pVtxSrc2);
 
@@ -2320,8 +2324,8 @@ void FetchJit::Shuffle8bpcGatherd2(Shuffle8bpcArgs &args)
 
         // SIMD16 PSHUFB isnt part of AVX-512F, so split into SIMD8 for the sake of KNL, for now..
 
-        Value *vGatherResult_lo = EXTRACT2(vGatherResult, 0);
-        Value *vGatherResult_hi = EXTRACT2(vGatherResult, 1);
+        Value *vGatherResult_lo = EXTRACT_16(vGatherResult, 0);
+        Value *vGatherResult_hi = EXTRACT_16(vGatherResult, 1);
 
         Value *vShufResult_lo = BITCAST(PSHUFB(BITCAST(vGatherResult_lo, v32x8Ty), vConstMask), vGatherTy);
         Value *vShufResult_hi = BITCAST(PSHUFB(BITCAST(vGatherResult_hi, v32x8Ty), vConstMask), vGatherTy);
@@ -2399,7 +2403,7 @@ void FetchJit::Shuffle8bpcGatherd2(Shuffle8bpcArgs &args)
                         temp_hi = FMUL(CAST(fpCast, temp_hi, mSimdFP32Ty), conversionFactor);
                     }
 
-                    vVertexElements[currentVertexElement] = JOIN2(temp_lo, temp_hi);
+                    vVertexElements[currentVertexElement] = JOIN_16(temp_lo, temp_hi);
 
                     currentVertexElement += 1;
                 }
@@ -2480,8 +2484,8 @@ void FetchJit::Shuffle8bpcGatherd2(Shuffle8bpcArgs &args)
                         break;
                     }
 
-                    Value *vGatherResult_lo = EXTRACT2(vGatherResult, 0);
-                    Value *vGatherResult_hi = EXTRACT2(vGatherResult, 1);
+                    Value *vGatherResult_lo = EXTRACT_16(vGatherResult, 0);
+                    Value *vGatherResult_hi = EXTRACT_16(vGatherResult, 1);
 
                     Value *temp_lo = BITCAST(PSHUFB(BITCAST(vGatherResult_lo, v32x8Ty), vConstMask), vGatherTy);
                     Value *temp_hi = BITCAST(PSHUFB(BITCAST(vGatherResult_hi, v32x8Ty), vConstMask), vGatherTy);
@@ -2497,7 +2501,7 @@ void FetchJit::Shuffle8bpcGatherd2(Shuffle8bpcArgs &args)
                         temp_hi = FMUL(CAST(fpCast, temp_hi, mSimdFP32Ty), conversionFactor);
                     }
 
-                    vVertexElements[currentVertexElement] = JOIN2(temp_lo, temp_hi);
+                    vVertexElements[currentVertexElement] = JOIN_16(temp_lo, temp_hi);
 
                     currentVertexElement += 1;
                 }
@@ -2785,8 +2789,8 @@ void FetchJit::Shuffle16bpcGather2(Shuffle16bpcArgs &args)
         {
             // SIMD16 PSHUFB isnt part of AVX-512F, so split into SIMD8 for the sake of KNL, for now..
 
-            Value *vGatherResult_lo = EXTRACT2(vGatherResult[0], 0);
-            Value *vGatherResult_hi = EXTRACT2(vGatherResult[0], 1);
+            Value *vGatherResult_lo = EXTRACT_16(vGatherResult[0], 0);
+            Value *vGatherResult_hi = EXTRACT_16(vGatherResult[0], 1);
 
             Value *vShufResult_lo = BITCAST(PSHUFB(BITCAST(vGatherResult_lo, v32x8Ty), vConstMask), vGatherTy);
             Value *vShufResult_hi = BITCAST(PSHUFB(BITCAST(vGatherResult_hi, v32x8Ty), vConstMask), vGatherTy);
@@ -2802,7 +2806,7 @@ void FetchJit::Shuffle16bpcGather2(Shuffle16bpcArgs &args)
             // 256i - 0    1    2    3    4    5    6    7
             //        xxxx xxxx xxxx xxxx yyyy yyyy yyyy yyyy
 #if 0
-            vi128XY = JOIN2(vi128XY_lo, vi128XY_hi);
+            vi128XY = JOIN_16(vi128XY_lo, vi128XY_hi);
 #endif
         }
 
@@ -2812,8 +2816,8 @@ void FetchJit::Shuffle16bpcGather2(Shuffle16bpcArgs &args)
         Value *vi128ZW_hi = nullptr;
         if (isComponentEnabled(compMask, 2) || isComponentEnabled(compMask, 3))
         {
-            Value *vGatherResult_lo = EXTRACT2(vGatherResult[1], 0);
-            Value *vGatherResult_hi = EXTRACT2(vGatherResult[1], 1);
+            Value *vGatherResult_lo = EXTRACT_16(vGatherResult[1], 0);
+            Value *vGatherResult_hi = EXTRACT_16(vGatherResult[1], 1);
 
             Value *vShufResult_lo = BITCAST(PSHUFB(BITCAST(vGatherResult_lo, v32x8Ty), vConstMask), vGatherTy);
             Value *vShufResult_hi = BITCAST(PSHUFB(BITCAST(vGatherResult_hi, v32x8Ty), vConstMask), vGatherTy);
@@ -2821,7 +2825,7 @@ void FetchJit::Shuffle16bpcGather2(Shuffle16bpcArgs &args)
             vi128ZW_lo = BITCAST(PERMD(vShufResult_lo, C<int32_t>({ 0, 1, 4, 5, 2, 3, 6, 7 })), v128bitTy);
             vi128ZW_hi = BITCAST(PERMD(vShufResult_hi, C<int32_t>({ 0, 1, 4, 5, 2, 3, 6, 7 })), v128bitTy);
 #if 0
-            vi128ZW = JOIN2(vi128ZW_lo, vi128ZW_hi);
+            vi128ZW = JOIN_16(vi128ZW_lo, vi128ZW_hi);
 #endif
         }
 
@@ -2868,7 +2872,7 @@ void FetchJit::Shuffle16bpcGather2(Shuffle16bpcArgs &args)
                         Value *temp_lo = CVTPH2PS(BITCAST(VEXTRACT(selectedPermute_lo, C(lane)), v8x16Ty));
                         Value *temp_hi = CVTPH2PS(BITCAST(VEXTRACT(selectedPermute_hi, C(lane)), v8x16Ty));
 
-                        vVertexElements[currentVertexElement] = JOIN2(temp_lo, temp_hi);
+                        vVertexElements[currentVertexElement] = JOIN_16(temp_lo, temp_hi);
                     }
                     else
                     {
@@ -2883,7 +2887,7 @@ void FetchJit::Shuffle16bpcGather2(Shuffle16bpcArgs &args)
                             temp_hi = FMUL(CAST(IntToFpCast, temp_hi, mSimdFP32Ty), conversionFactor);
                         }
 
-                        vVertexElements[currentVertexElement] = JOIN2(temp_lo, temp_hi);
+                        vVertexElements[currentVertexElement] = JOIN_16(temp_lo, temp_hi);
                     }
 
                     currentVertexElement += 1;
@@ -2960,8 +2964,8 @@ void FetchJit::Shuffle16bpcGather2(Shuffle16bpcArgs &args)
 
                     // SIMD16 PSHUFB isnt part of AVX-512F, so split into SIMD8 for the sake of KNL, for now..
 
-                    Value *vGatherResult_lo = EXTRACT2(vGatherResult[selectedGather], 0);
-                    Value *vGatherResult_hi = EXTRACT2(vGatherResult[selectedGather], 1);
+                    Value *vGatherResult_lo = EXTRACT_16(vGatherResult[selectedGather], 0);
+                    Value *vGatherResult_hi = EXTRACT_16(vGatherResult[selectedGather], 1);
 
                     Value *temp_lo = BITCAST(PSHUFB(BITCAST(vGatherResult_lo, v32x8Ty), vConstMask[selectedMask]), vGatherTy);
                     Value *temp_hi = BITCAST(PSHUFB(BITCAST(vGatherResult_hi, v32x8Ty), vConstMask[selectedMask]), vGatherTy);
@@ -2977,7 +2981,7 @@ void FetchJit::Shuffle16bpcGather2(Shuffle16bpcArgs &args)
                         temp_hi = FMUL(CAST(fpCast, temp_hi, mSimdFP32Ty), conversionFactor);
                     }
 
-                    vVertexElements[currentVertexElement] = JOIN2(temp_lo, temp_hi);
+                    vVertexElements[currentVertexElement] = JOIN_16(temp_lo, temp_hi);
 
                     currentVertexElement += 1;
                 }
@@ -3052,7 +3056,7 @@ void FetchJit::StoreVertexElements2(Value* pVtxOut, const uint32_t outputElt, co
 #if FETCH_DUMP_VERTEX
             PRINT("vVertexElements[%d]: 0x%x\n", { C(c), vVertexElements[c] });
 #endif
-            vVertexElements[c] = BITCAST(vVertexElements[c], mSimd2FP32Ty);
+            vVertexElements[c] = BITCAST(vVertexElements[c], mSimd16FP32Ty);
         }
 #if FETCH_DUMP_VERTEX
         else
@@ -3080,14 +3084,18 @@ Value* FetchJit::GenerateCompCtrlVector(const ComponentControl ctrl)
 {
     switch(ctrl)
     {
-        case NoStore:   return VUNDEF_I();
-        case Store0:    return VIMMED1(0);
-        case Store1Fp:  return VIMMED1(1.0f);
-        case Store1Int: return VIMMED1(1);
+        case NoStore:   
+            return VUNDEF_I();
+        case Store0:    
+            return VIMMED1(0);
+        case Store1Fp:  
+            return VIMMED1(1.0f);
+        case Store1Int: 
+            return VIMMED1(1);
         case StoreVertexId:
         {
 #if USE_SIMD16_SHADERS
-            Value* pId;
+            Value *pId;
             if (useVertexID2)
             {
                 pId = BITCAST(LOAD(GEP(mpFetchInfo, { 0, SWR_FETCH_CONTEXT_VertexID2 })), mSimdFP32Ty);
@@ -3097,17 +3105,18 @@ Value* FetchJit::GenerateCompCtrlVector(const ComponentControl ctrl)
                 pId = BITCAST(LOAD(GEP(mpFetchInfo, { 0, SWR_FETCH_CONTEXT_VertexID })), mSimdFP32Ty);
             }
 #else
-            Value* pId = BITCAST(LOAD(GEP(mpFetchInfo, { 0, SWR_FETCH_CONTEXT_VertexID })), mSimdFP32Ty);
+            Value *pId = BITCAST(LOAD(GEP(mpFetchInfo, { 0, SWR_FETCH_CONTEXT_VertexID })), mSimdFP32Ty);
 #endif
             return pId;
         }
         case StoreInstanceId:
         {
-            Value* pId = BITCAST(LOAD(GEP(mpFetchInfo, { 0, SWR_FETCH_CONTEXT_CurInstance })), mFP32Ty);
+            Value *pId = BITCAST(LOAD(GEP(mpFetchInfo, { 0, SWR_FETCH_CONTEXT_CurInstance })), mFP32Ty);
             return VBROADCAST(pId);
         }
         case StoreSrc:
-        default:        SWR_INVALID("Invalid component control"); return VUNDEF_I();
+        default:        
+            SWR_INVALID("Invalid component control"); return VUNDEF_I();
     }
 }
 
@@ -3116,26 +3125,31 @@ Value* FetchJit::GenerateCompCtrlVector2(const ComponentControl ctrl)
 {
     switch (ctrl)
     {
-        case NoStore:   return VUNDEF2_I();
-        case Store0:    return VIMMED2_1(0);
-        case Store1Fp:  return VIMMED2_1(1.0f);
-        case Store1Int: return VIMMED2_1(1);
+        case NoStore:   
+            return VUNDEF_I_16();
+        case Store0:    
+            return VIMMED1_16(0);
+        case Store1Fp:  
+            return VIMMED1_16(1.0f);
+        case Store1Int: 
+            return VIMMED1_16(1);
         case StoreVertexId:
         {
-            Value* pId_lo = BITCAST(LOAD(GEP(mpFetchInfo, { 0, SWR_FETCH_CONTEXT_VertexID  })), mSimdFP32Ty);
-            Value* pId_hi = BITCAST(LOAD(GEP(mpFetchInfo, { 0, SWR_FETCH_CONTEXT_VertexID2 })), mSimdFP32Ty);
+            Value *pId_lo = BITCAST(LOAD(GEP(mpFetchInfo, { 0, SWR_FETCH_CONTEXT_VertexID  })), mSimdFP32Ty);
+            Value *pId_hi = BITCAST(LOAD(GEP(mpFetchInfo, { 0, SWR_FETCH_CONTEXT_VertexID2 })), mSimdFP32Ty);
 
-            Value *pId = JOIN2(pId_lo, pId_hi);
+            Value *pId = JOIN_16(pId_lo, pId_hi);
 
             return pId;
         }
         case StoreInstanceId:
         {
-            Value* pId = BITCAST(LOAD(GEP(mpFetchInfo, { 0, SWR_FETCH_CONTEXT_CurInstance })), mFP32Ty);
-            return VBROADCAST2(pId);
+            Value *pId = BITCAST(LOAD(GEP(mpFetchInfo, { 0, SWR_FETCH_CONTEXT_CurInstance })), mFP32Ty);
+            return VBROADCAST_16(pId);
         }
         case StoreSrc:
-        default:        SWR_INVALID("Invalid component control"); return VUNDEF2_I();
+        default:        
+            SWR_INVALID("Invalid component control"); return VUNDEF_I_16();
     }
 }