swr: [rasterizer core/common/jitter] gl_double support

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99214
Reviewed-by: Bruce Cherniak <[email protected]>

9 files changed, 341 insertions, 33 deletions

diff --git a/src/gallium/drivers/swr/rasterizer/common/formats.cpp b/src/gallium/drivers/swr/rasterizer/common/formats.cpp
index aba4c3f3380..72020ee1936 100644
--- a/src/gallium/drivers/swr/rasterizer/common/formats.cpp
+++ b/src/gallium/drivers/swr/rasterizer/common/formats.cpp
@@ -149,16 +149,26 @@ const SWR_FORMAT_INFO gFormatInfo[] = {
         { 0.0f, 0.0f, 0.0f, 0.0f },
         1, 1
     },
-    // padding (0x5)
+    // R64G64_FLOAT (0x5)
     {
-        nullptr,
-        { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN },
-        { 0, 0, 0, 0 },{ 0, 0, 0, 0 },{ 0, 0, 0, 0 },
-        0, 0, 0, false, false, false, false,
-        { false, false, false, false },
-        { 0.0f, 0.0f, 0.0f, 0.0f },
-        1, 1
+        "R64G64_FLOAT",
+        { SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN },
+        { 0, 0, 0, 0x3f800000 }, // Defaults for missing components
+        { 0, 1, 0, 0 }, // Swizzle
+        { 64, 64, 0, 0 }, // Bits per component
+        128, // Bits per element
+        16, // Bytes per element
+        2, // Num components
+        false, // isSRGB
+        false, // isBC
+        false, // isSubsampled
+        false, // isLuminance
+        { false, false, false, false }, // Is normalized?
+        { 1.0f, 1.0f, 0, 0 }, // To float scale factor
+        1, // bcWidth
+        1, // bcHeight
     },
+
     // R32G32B32X32_FLOAT (0x6)
     {
         "R32G32B32X32_FLOAT",
@@ -1719,16 +1729,26 @@ const SWR_FORMAT_INFO gFormatInfo[] = {
         { 0.0f, 0.0f, 0.0f, 0.0f },
         1, 1
     },
-    // padding (0x8D)
+    // R64_FLOAT (0x8D)
     {
-        nullptr,
-        { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN },
-        { 0, 0, 0, 0 },{ 0, 0, 0, 0 },{ 0, 0, 0, 0 },
-        0, 0, 0, false, false, false, false,
-        { false, false, false, false },
-        { 0.0f, 0.0f, 0.0f, 0.0f },
-        1, 1
+        "R64_FLOAT",
+        { SWR_TYPE_FLOAT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN },
+        { 0, 0, 0, 0x3f800000 }, // Defaults for missing components
+        { 0, 0, 0, 0 }, // Swizzle
+        { 64, 0, 0, 0 }, // Bits per component
+        64, // Bits per element
+        8, // Bytes per element
+        1, // Num components
+        false, // isSRGB
+        false, // isBC
+        false, // isSubsampled
+        false, // isLuminance
+        { false, false, false, false }, // Is normalized?
+        { 1.0f, 0, 0, 0 }, // To float scale factor
+        1, // bcWidth
+        1, // bcHeight
     },
+
     // R16G16B16X16_UNORM (0x8E)
     {
         "R16G16B16X16_UNORM",
@@ -5529,26 +5549,46 @@ const SWR_FORMAT_INFO gFormatInfo[] = {
         1, // bcHeight
     },
 
-    // padding (0x197)
+    // R64G64B64A64_FLOAT (0x197)
     {
-        nullptr,
-        { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN },
-        { 0, 0, 0, 0 },{ 0, 0, 0, 0 },{ 0, 0, 0, 0 },
-        0, 0, 0, false, false, false, false,
-        { false, false, false, false },
-        { 0.0f, 0.0f, 0.0f, 0.0f },
-        1, 1
+        "R64G64B64A64_FLOAT",
+        { SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_FLOAT },
+        { 0, 0, 0, 0x3f800000 }, // Defaults for missing components
+        { 0, 1, 2, 3 }, // Swizzle
+        { 64, 64, 64, 64 }, // Bits per component
+        256, // Bits per element
+        32, // Bytes per element
+        4, // Num components
+        false, // isSRGB
+        false, // isBC
+        false, // isSubsampled
+        false, // isLuminance
+        { false, false, false, false }, // Is normalized?
+        { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor
+        1, // bcWidth
+        1, // bcHeight
     },
-    // padding (0x198)
+
+    // R64G64B64_FLOAT (0x198)
     {
-        nullptr,
-        { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN },
-        { 0, 0, 0, 0 },{ 0, 0, 0, 0 },{ 0, 0, 0, 0 },
-        0, 0, 0, false, false, false, false,
-        { false, false, false, false },
-        { 0.0f, 0.0f, 0.0f, 0.0f },
-        1, 1
+        "R64G64B64_FLOAT",
+        { SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_UNKNOWN },
+        { 0, 0, 0, 0x3f800000 }, // Defaults for missing components
+        { 0, 1, 2, 0 }, // Swizzle
+        { 64, 64, 64, 0 }, // Bits per component
+        192, // Bits per element
+        24, // Bytes per element
+        3, // Num components
+        false, // isSRGB
+        false, // isBC
+        false, // isSubsampled
+        false, // isLuminance
+        { false, false, false, false }, // Is normalized?
+        { 1.0f, 1.0f, 1.0f, 0 }, // To float scale factor
+        1, // bcWidth
+        1, // bcHeight
     },
+
     // BC4_SNORM (0x199)
     {
         "BC4_SNORM",
diff --git a/src/gallium/drivers/swr/rasterizer/common/formats.h b/src/gallium/drivers/swr/rasterizer/common/formats.h
index 3d686d3e0b0..0056a56f829 100644
--- a/src/gallium/drivers/swr/rasterizer/common/formats.h
+++ b/src/gallium/drivers/swr/rasterizer/common/formats.h
@@ -57,6 +57,7 @@ enum SWR_FORMAT
     R32G32B32A32_FLOAT          = 0x0,
     R32G32B32A32_SINT           = 0x1,
     R32G32B32A32_UINT           = 0x2,
+    R64G64_FLOAT                = 0x5,
     R32G32B32X32_FLOAT          = 0x6,
     R32G32B32A32_SSCALED        = 0x7,
     R32G32B32A32_USCALED        = 0x8,
@@ -78,6 +79,7 @@ enum SWR_FORMAT
     R32_FLOAT_X8X24_TYPELESS    = 0x88,
     X32_TYPELESS_G8X24_UINT     = 0x89,
     L32A32_FLOAT                = 0x8A,
+    R64_FLOAT                   = 0x8D,
     R16G16B16X16_UNORM          = 0x8E,
     R16G16B16X16_FLOAT          = 0x8F,
     L32X32_FLOAT                = 0x91,
@@ -193,6 +195,8 @@ enum SWR_FORMAT
     R8G8B8_SNORM                = 0x194,
     R8G8B8_SSCALED              = 0x195,
     R8G8B8_USCALED              = 0x196,
+    R64G64B64A64_FLOAT          = 0x197,
+    R64G64B64_FLOAT             = 0x198,
     BC4_SNORM                   = 0x199,
     BC5_SNORM                   = 0x19A,
     R16G16B16_FLOAT             = 0x19B,
diff --git a/src/gallium/drivers/swr/rasterizer/core/format_traits.h b/src/gallium/drivers/swr/rasterizer/core/format_traits.h
index 59d4e7d7e9f..6c428043878 100644
--- a/src/gallium/drivers/swr/rasterizer/core/format_traits.h
+++ b/src/gallium/drivers/swr/rasterizer/core/format_traits.h
@@ -134,6 +134,28 @@ template<> struct FormatTraits<R32G32B32A32_UINT> :
 };
 
 //////////////////////////////////////////////////////////////////////////
+/// FormatTraits<R64G64_FLOAT> - Format traits specialization for R64G64_FLOAT
+//////////////////////////////////////////////////////////////////////////
+template<> struct FormatTraits<R64G64_FLOAT> :
+    ComponentTraits<SWR_TYPE_FLOAT, 64, SWR_TYPE_FLOAT, 64>,
+    FormatSwizzle<0, 1>,
+    Defaults<0, 0, 0, 0x3f800000>
+{
+    static const uint32_t bpp{ 128 };
+    static const uint32_t numComps{ 2 };
+    static const bool hasAlpha{ false };
+    static const uint32_t alphaComp{ 0 };
+    static const bool isSRGB{ false };
+    static const bool isBC{ false };
+    static const bool isSubsampled{ false };
+    static const uint32_t bcWidth{ 1 };
+    static const uint32_t bcHeight{ 1 };
+
+    typedef Transpose64_64  TransposeT;
+    typedef Format2<64, 64> FormatT;
+};
+
+//////////////////////////////////////////////////////////////////////////
 /// FormatTraits<R32G32B32X32_FLOAT> - Format traits specialization for R32G32B32X32_FLOAT
 //////////////////////////////////////////////////////////////////////////
 template<> struct FormatTraits<R32G32B32X32_FLOAT> :
@@ -596,6 +618,28 @@ template<> struct FormatTraits<L32A32_FLOAT> :
 };
 
 //////////////////////////////////////////////////////////////////////////
+/// FormatTraits<R64_FLOAT> - Format traits specialization for R64_FLOAT
+//////////////////////////////////////////////////////////////////////////
+template<> struct FormatTraits<R64_FLOAT> :
+    ComponentTraits<SWR_TYPE_FLOAT, 64>,
+    FormatSwizzle<0>,
+    Defaults<0, 0, 0, 0x3f800000>
+{
+    static const uint32_t bpp{ 64 };
+    static const uint32_t numComps{ 1 };
+    static const bool hasAlpha{ false };
+    static const uint32_t alphaComp{ 0 };
+    static const bool isSRGB{ false };
+    static const bool isBC{ false };
+    static const bool isSubsampled{ false };
+    static const uint32_t bcWidth{ 1 };
+    static const uint32_t bcHeight{ 1 };
+
+    typedef TransposeSingleComponent<64> TransposeT;
+    typedef Format1<64>                  FormatT;
+};
+
+//////////////////////////////////////////////////////////////////////////
 /// FormatTraits<R16G16B16X16_UNORM> - Format traits specialization for R16G16B16X16_UNORM
 //////////////////////////////////////////////////////////////////////////
 template<> struct FormatTraits<R16G16B16X16_UNORM> :
@@ -3126,6 +3170,50 @@ template<> struct FormatTraits<R8G8B8_USCALED> :
 };
 
 //////////////////////////////////////////////////////////////////////////
+/// FormatTraits<R64G64B64A64_FLOAT> - Format traits specialization for R64G64B64A64_FLOAT
+//////////////////////////////////////////////////////////////////////////
+template<> struct FormatTraits<R64G64B64A64_FLOAT> :
+    ComponentTraits<SWR_TYPE_FLOAT, 64, SWR_TYPE_FLOAT, 64, SWR_TYPE_FLOAT, 64, SWR_TYPE_FLOAT, 64>,
+    FormatSwizzle<0, 1, 2, 3>,
+    Defaults<0, 0, 0, 0x3f800000>
+{
+    static const uint32_t bpp{ 256 };
+    static const uint32_t numComps{ 4 };
+    static const bool hasAlpha{ true };
+    static const uint32_t alphaComp{ 3 };
+    static const bool isSRGB{ false };
+    static const bool isBC{ false };
+    static const bool isSubsampled{ false };
+    static const uint32_t bcWidth{ 1 };
+    static const uint32_t bcHeight{ 1 };
+
+    typedef Transpose64_64_64_64    TransposeT;
+    typedef Format4<64, 64, 64, 64> FormatT;
+};
+
+//////////////////////////////////////////////////////////////////////////
+/// FormatTraits<R64G64B64_FLOAT> - Format traits specialization for R64G64B64_FLOAT
+//////////////////////////////////////////////////////////////////////////
+template<> struct FormatTraits<R64G64B64_FLOAT> :
+    ComponentTraits<SWR_TYPE_FLOAT, 64, SWR_TYPE_FLOAT, 64, SWR_TYPE_FLOAT, 64>,
+    FormatSwizzle<0, 1, 2>,
+    Defaults<0, 0, 0, 0x3f800000>
+{
+    static const uint32_t bpp{ 192 };
+    static const uint32_t numComps{ 3 };
+    static const bool hasAlpha{ false };
+    static const uint32_t alphaComp{ 0 };
+    static const bool isSRGB{ false };
+    static const bool isBC{ false };
+    static const bool isSubsampled{ false };
+    static const uint32_t bcWidth{ 1 };
+    static const uint32_t bcHeight{ 1 };
+
+    typedef Transpose64_64_64   TransposeT;
+    typedef Format3<64, 64, 64> FormatT;
+};
+
+//////////////////////////////////////////////////////////////////////////
 /// FormatTraits<BC4_SNORM> - Format traits specialization for BC4_SNORM
 //////////////////////////////////////////////////////////////////////////
 template<> struct FormatTraits<BC4_SNORM> :
diff --git a/src/gallium/drivers/swr/rasterizer/core/utils.h b/src/gallium/drivers/swr/rasterizer/core/utils.h
index a2365757a7f..0e2cb47f2bd 100644
--- a/src/gallium/drivers/swr/rasterizer/core/utils.h
+++ b/src/gallium/drivers/swr/rasterizer/core/utils.h
@@ -856,6 +856,70 @@ struct Transpose11_11_10
 #endif
 };
 
+//////////////////////////////////////////////////////////////////////////
+/// Transpose64
+//////////////////////////////////////////////////////////////////////////
+struct Transpose64
+{
+    //////////////////////////////////////////////////////////////////////////
+    /// @brief Performs an SOA to AOS conversion
+    /// @param pSrc - source data in SOA form
+    /// @param pDst - output data in AOS form
+    static void Transpose(const uint8_t* pSrc, uint8_t* pDst) = delete;
+#if ENABLE_AVX512_SIMD16
+
+    static void Transpose_16(const uint8_t* pSrc, uint8_t* pDst) = delete;
+#endif
+};
+
+//////////////////////////////////////////////////////////////////////////
+/// Transpose64_64
+//////////////////////////////////////////////////////////////////////////
+struct Transpose64_64
+{
+    //////////////////////////////////////////////////////////////////////////
+    /// @brief Performs an SOA to AOS conversion
+    /// @param pSrc - source data in SOA form
+    /// @param pDst - output data in AOS form
+    static void Transpose(const uint8_t* pSrc, uint8_t* pDst) = delete;
+#if ENABLE_AVX512_SIMD16
+
+    static void Transpose_16(const uint8_t* pSrc, uint8_t* pDst) = delete;
+#endif
+};
+
+//////////////////////////////////////////////////////////////////////////
+/// Transpose64_64_64
+//////////////////////////////////////////////////////////////////////////
+struct Transpose64_64_64
+{
+    //////////////////////////////////////////////////////////////////////////
+    /// @brief Performs an SOA to AOS conversion
+    /// @param pSrc - source data in SOA form
+    /// @param pDst - output data in AOS form
+    static void Transpose(const uint8_t* pSrc, uint8_t* pDst) = delete;
+#if ENABLE_AVX512_SIMD16
+
+    static void Transpose_16(const uint8_t* pSrc, uint8_t* pDst) = delete;
+#endif
+};
+
+//////////////////////////////////////////////////////////////////////////
+/// Transpose64_64_64_64
+//////////////////////////////////////////////////////////////////////////
+struct Transpose64_64_64_64
+{
+    //////////////////////////////////////////////////////////////////////////
+    /// @brief Performs an SOA to AOS conversion
+    /// @param pSrc - source data in SOA form
+    /// @param pDst - output data in AOS form
+    static void Transpose(const uint8_t* pSrc, uint8_t* pDst) = delete;
+#if ENABLE_AVX512_SIMD16
+
+    static void Transpose_16(const uint8_t* pSrc, uint8_t* pDst) = delete;
+#endif
+};
+
 // helper function to unroll loops
 template<int Begin, int End, int Step = 1>
 struct UnrollerL {
diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp b/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp
index 8120a2f21ed..8744eb6554d 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp
+++ b/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp
@@ -632,6 +632,55 @@ namespace SwrJit
     }
 
     //////////////////////////////////////////////////////////////////////////
+    /// @brief Generate a masked gather operation in LLVM IR.  If not
+    /// supported on the underlying platform, emulate it with loads
+    /// @param vSrc - SIMD wide value that will be loaded if mask is invalid
+    /// @param pBase - Int8* base VB address pointer value
+    /// @param vIndices - SIMD wide value of VB byte offsets
+    /// @param vMask - SIMD wide mask that controls whether to access memory or the src values
+    /// @param scale - value to scale indices by
+    Value *Builder::GATHERPD(Value* vSrc, Value* pBase, Value* vIndices, Value* vMask, Value* scale)
+    {
+        Value* vGather;
+
+        // use avx2 gather instruction if available
+        if(JM()->mArch.AVX2())
+        {
+            vGather = VGATHERPD(vSrc, pBase, vIndices, vMask, scale);
+        }
+        else
+        {
+            Value* pStack = STACKSAVE();
+
+            // store vSrc on the stack.  this way we can select between a valid load address and the vSrc address
+            Value* vSrcPtr = ALLOCA(vSrc->getType());
+            STORE(vSrc, vSrcPtr);
+
+            vGather = UndefValue::get(VectorType::get(mDoubleTy, 4));
+            Value *vScaleVec = VECTOR_SPLAT(4, Z_EXT(scale,mInt32Ty));
+            Value *vOffsets = MUL(vIndices,vScaleVec);
+            Value *mask = MASK(vMask);
+            for(uint32_t i = 0; i < mVWidth/2; ++i)
+            {
+                // single component byte index
+                Value *offset = VEXTRACT(vOffsets,C(i));
+                // byte pointer to component
+                Value *loadAddress = GEP(pBase,offset);
+                loadAddress = BITCAST(loadAddress,PointerType::get(mDoubleTy,0));
+                // pointer to the value to load if we're masking off a component
+                Value *maskLoadAddress = GEP(vSrcPtr,{C(0), C(i)});
+                Value *selMask = VEXTRACT(mask,C(i));
+                // switch in a safe address to load if we're trying to access a vertex
+                Value *validAddress = SELECT(selMask, loadAddress, maskLoadAddress);
+                Value *val = LOAD(validAddress);
+                vGather = VINSERT(vGather,val,C(i));
+            }
+            STACKRESTORE(pStack);
+        }
+        return vGather;
+    }
+
+    //////////////////////////////////////////////////////////////////////////
     /// @brief convert x86 <N x float> mask to llvm <N x i1> mask
     Value* Builder::MASK(Value* vmask)
     {
diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.h b/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.h
index 86f7bf21a6b..67f938e192e 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.h
+++ b/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.h
@@ -113,6 +113,8 @@ Value *GATHERDD(Value* src, Value* pBase, Value* indices, Value* mask, Value* sc
 void GATHER4DD(const SWR_FORMAT_INFO &info, Value* pSrcBase, Value* byteOffsets,
                Value* mask, Value* vGatherComponents[], bool bPackedOutput);
 
+Value *GATHERPD(Value* src, Value* pBase, Value* indices, Value* mask, Value* scale);
+
 void SCATTERPS(Value* pDst, Value* vSrc, Value* vOffsets, Value* vMask);
 
 void Shuffle8bpcGather4(const SWR_FORMAT_INFO &info, Value* vGatherInput, Value* vGatherOutput[], 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 c5936e59bfe..984aab67cd6 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp
+++ b/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp
@@ -519,7 +519,7 @@ void FetchJit::JitLoadVertices(const FETCH_COMPILE_STATE &fetchState, Value* str
 bool FetchJit::IsOddFormat(SWR_FORMAT format)
 {
     const SWR_FORMAT_INFO& info = GetFormatInfo(format);
-    if (info.bpc[0] != 8 && info.bpc[0] != 16 && info.bpc[0] != 32)
+    if (info.bpc[0] != 8 && info.bpc[0] != 16 && info.bpc[0] != 32 && info.bpc[0] != 64)
     {
         return true;
     }
@@ -914,6 +914,58 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
                     }
                 }
                     break;
+                case 64:
+                {
+                    for (uint32_t i = 0; i < 4; i++)
+                    {
+                        if (isComponentEnabled(compMask, i))
+                        {
+                            // if we need to gather the component
+                            if (compCtrl[i] == StoreSrc)
+                            {
+                                Value *vMaskLo = VSHUFFLE(pMask, VUNDEF(mInt1Ty, 8), C({0, 1, 2, 3}));
+                                Value *vMaskHi = VSHUFFLE(pMask, VUNDEF(mInt1Ty, 8), C({4, 5, 6, 7}));
+                                vMaskLo = S_EXT(vMaskLo, VectorType::get(mInt64Ty, 4));
+                                vMaskHi = S_EXT(vMaskHi, VectorType::get(mInt64Ty, 4));
+                                vMaskLo = BITCAST(vMaskLo, VectorType::get(mDoubleTy, 4));
+                                vMaskHi = BITCAST(vMaskHi, VectorType::get(mDoubleTy, 4));
+
+                                Value *vOffsetsLo = VEXTRACTI128(vOffsets, C(0));
+                                Value *vOffsetsHi = VEXTRACTI128(vOffsets, C(1));
+
+                                Value *vZeroDouble = VECTOR_SPLAT(4, ConstantFP::get(IRB()->getDoubleTy(), 0.0f));
+
+                                Value* pGatherLo = GATHERPD(vZeroDouble,
+                                                            pStreamBase, vOffsetsLo, vMaskLo, C((char)1));
+                                Value* pGatherHi = GATHERPD(vZeroDouble,
+                                                            pStreamBase, vOffsetsHi, vMaskHi, C((char)1));
+
+                                pGatherLo = VCVTPD2PS(pGatherLo);
+                                pGatherHi = VCVTPD2PS(pGatherHi);
+
+                                Value *pGather = VSHUFFLE(pGatherLo, pGatherHi, C({0, 1, 2, 3, 4, 5, 6, 7}));
+
+                                vVertexElements[currentVertexElement++] = pGather;
+                            }
+                            else
+                            {
+                                vVertexElements[currentVertexElement++] = GenerateCompCtrlVector(compCtrl[i]);
+                            }
+
+                            if (currentVertexElement > 3)
+                            {
+                                StoreVertexElements(pVtxOut, outputElt++, 4, vVertexElements);
+                                // reset to the next vVertexElement to output
+                                currentVertexElement = 0;
+                            }
+
+                        }
+
+                        // offset base to the next component  in the vertex to gather
+                        pStreamBase = GEP(pStreamBase, C((char)8));
+                    }
+                }
+                    break;
                 default:
                     SWR_ASSERT(0, "Tried to fetch invalid FP format");
                     break;
@@ -1730,6 +1782,8 @@ PFN_FETCH_FUNC JitFetchFunc(HANDLE hJitMgr, const HANDLE hFunc)
     fclose(fd);
 #endif
 
+    pJitMgr->DumpAsm(const_cast<llvm::Function*>(func), "final");
+
     return pfnFetch;
 }
 
diff --git a/src/gallium/drivers/swr/rasterizer/jitter/scripts/gen_llvm_ir_macros.py b/src/gallium/drivers/swr/rasterizer/jitter/scripts/gen_llvm_ir_macros.py
index c6d09413211..70d35762bc5 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/scripts/gen_llvm_ir_macros.py
+++ b/src/gallium/drivers/swr/rasterizer/jitter/scripts/gen_llvm_ir_macros.py
@@ -84,6 +84,7 @@ inst_aliases = {
 }
 
 intrinsics = [
+        ["VGATHERPD", "x86_avx2_gather_d_pd_256", ["src", "pBase", "indices", "mask", "scale"]],
         ["VGATHERPS", "x86_avx2_gather_d_ps_256", ["src", "pBase", "indices", "mask", "scale"]],
         ["VGATHERDD", "x86_avx2_gather_d_d_256", ["src", "pBase", "indices", "mask", "scale"]],
         ["VSQRTPS", "x86_avx_sqrt_ps_256", ["a"]],
@@ -101,6 +102,7 @@ intrinsics = [
         ["VPSHUFB", "x86_avx2_pshuf_b", ["a", "b"]],
         ["VPERMD", "x86_avx2_permd", ["a", "idx"]],
         ["VPERMPS", "x86_avx2_permps", ["idx", "a"]],
+        ["VCVTPD2PS", "x86_avx_cvt_pd2_ps_256", ["a"]],
         ["VCVTPH2PS", "x86_vcvtph2ps_256", ["a"]],
         ["VCVTPS2PH", "x86_vcvtps2ph_256", ["a", "round"]],
         ["VHSUBPS", "x86_avx_hsub_ps_256", ["a", "b"]],
diff --git a/src/gallium/drivers/swr/swr_screen.cpp b/src/gallium/drivers/swr/swr_screen.cpp
index 6ff21cdf186..cc8030eaff6 100644
--- a/src/gallium/drivers/swr/swr_screen.cpp
+++ b/src/gallium/drivers/swr/swr_screen.cpp
@@ -537,6 +537,11 @@ mesa_to_swr_format(enum pipe_format format)
       {PIPE_FORMAT_R32G32B32_FIXED,        R32G32B32_SFIXED},
       {PIPE_FORMAT_R32G32B32A32_FIXED,     R32G32B32A32_SFIXED},
 
+      {PIPE_FORMAT_R64_FLOAT,              R64_FLOAT},
+      {PIPE_FORMAT_R64G64_FLOAT,           R64G64_FLOAT},
+      {PIPE_FORMAT_R64G64B64_FLOAT,        R64G64B64_FLOAT},
+      {PIPE_FORMAT_R64G64B64A64_FLOAT,     R64G64B64A64_FLOAT},
+
       /* These formats have entries in SWR but don't have Load/StoreTile
        * implementations. That means these aren't renderable, and thus having
        * a mapping entry here is detrimental.