swr: [rasterizer common/core/jitter] fetch support for GL_FIXED

v2: use fmul(1/65536) instead of fdiv(65535)

Reviewed-by: Bruce Cherniak <[email protected]>

4 files changed, 179 insertions, 34 deletions

diff --git a/src/gallium/drivers/swr/rasterizer/common/formats.cpp b/src/gallium/drivers/swr/rasterizer/common/formats.cpp
index b3a95f2bf03..aba4c3f3380 100644
--- a/src/gallium/drivers/swr/rasterizer/common/formats.cpp
+++ b/src/gallium/drivers/swr/rasterizer/common/formats.cpp
@@ -449,16 +449,26 @@ const SWR_FORMAT_INFO gFormatInfo[] = {
         { 0.0f, 0.0f, 0.0f, 0.0f },
         1, 1
     },
-    // padding (0x20)
+    // R32G32B32A32_SFIXED (0x20)
     {
-        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
+        "R32G32B32A32_SFIXED",
+        { SWR_TYPE_SFIXED, SWR_TYPE_SFIXED, SWR_TYPE_SFIXED, SWR_TYPE_SFIXED },
+        { 0, 0, 0, 0x3f800000 }, // Defaults for missing components
+        { 0, 1, 2, 3 }, // Swizzle
+        { 32, 32, 32, 32 }, // Bits per component
+        128, // Bits per element
+        16, // 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 (0x21)
     {
         nullptr,
@@ -979,16 +989,26 @@ const SWR_FORMAT_INFO gFormatInfo[] = {
         { 0.0f, 0.0f, 0.0f, 0.0f },
         1, 1
     },
-    // padding (0x50)
+    // R32G32B32_SFIXED (0x50)
     {
-        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
+        "R32G32B32_SFIXED",
+        { SWR_TYPE_SFIXED, SWR_TYPE_SFIXED, SWR_TYPE_SFIXED, SWR_TYPE_UNKNOWN },
+        { 0, 0, 0, 0x3f800000 }, // Defaults for missing components
+        { 0, 1, 2, 0 }, // Swizzle
+        { 32, 32, 32, 0 }, // Bits per component
+        96, // Bits per element
+        12, // 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
     },
+
     // padding (0x51)
     {
         nullptr,
@@ -1969,16 +1989,26 @@ const SWR_FORMAT_INFO gFormatInfo[] = {
         { 0.0f, 0.0f, 0.0f, 0.0f },
         1, 1
     },
-    // padding (0xA0)
+    // R32G32_SFIXED (0xA0)
     {
-        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
+        "R32G32_SFIXED",
+        { SWR_TYPE_SFIXED, SWR_TYPE_SFIXED, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN },
+        { 0, 0, 0, 0x3f800000 }, // Defaults for missing components
+        { 0, 1, 0, 0 }, // Swizzle
+        { 32, 32, 0, 0 }, // Bits per component
+        64, // Bits per element
+        8, // 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
     },
+
     // padding (0xA1)
     {
         nullptr,
@@ -5909,16 +5939,26 @@ const SWR_FORMAT_INFO gFormatInfo[] = {
         1, // bcHeight
     },
 
-    // padding (0x1B2)
+    // R32_SFIXED (0x1B2)
     {
-        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
+        "R32_SFIXED",
+        { SWR_TYPE_SFIXED, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN },
+        { 0, 0, 0, 0x3f800000 }, // Defaults for missing components
+        { 0, 0, 0, 0 }, // Swizzle
+        { 32, 0, 0, 0 }, // Bits per component
+        32, // Bits per element
+        4, // 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
     },
+
     // R10G10B10A2_SNORM (0x1B3)
     {
         "R10G10B10A2_SNORM",
diff --git a/src/gallium/drivers/swr/rasterizer/common/formats.h b/src/gallium/drivers/swr/rasterizer/common/formats.h
index dd5b4999e39..3d686d3e0b0 100644
--- a/src/gallium/drivers/swr/rasterizer/common/formats.h
+++ b/src/gallium/drivers/swr/rasterizer/common/formats.h
@@ -20,7 +20,7 @@
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 * 
-* @file formats.h
+* @file gen_formats.h
 * 
 * @brief auto-generated file
 * 
@@ -46,6 +46,7 @@ enum SWR_TYPE
     SWR_TYPE_FLOAT,
     SWR_TYPE_SSCALED,
     SWR_TYPE_USCALED,
+    SWR_TYPE_SFIXED,
 };
 
 //////////////////////////////////////////////////////////////////////////
@@ -59,11 +60,13 @@ enum SWR_FORMAT
     R32G32B32X32_FLOAT          = 0x6,
     R32G32B32A32_SSCALED        = 0x7,
     R32G32B32A32_USCALED        = 0x8,
+    R32G32B32A32_SFIXED         = 0x20,
     R32G32B32_FLOAT             = 0x40,
     R32G32B32_SINT              = 0x41,
     R32G32B32_UINT              = 0x42,
     R32G32B32_SSCALED           = 0x45,
     R32G32B32_USCALED           = 0x46,
+    R32G32B32_SFIXED            = 0x50,
     R16G16B16A16_UNORM          = 0x80,
     R16G16B16A16_SNORM          = 0x81,
     R16G16B16A16_SINT           = 0x82,
@@ -83,6 +86,7 @@ enum SWR_FORMAT
     R16G16B16A16_USCALED        = 0x94,
     R32G32_SSCALED              = 0x95,
     R32G32_USCALED              = 0x96,
+    R32G32_SFIXED               = 0xA0,
     B8G8R8A8_UNORM              = 0xC0,
     B8G8R8A8_UNORM_SRGB         = 0xC1,
     R10G10B10A2_UNORM           = 0xC2,
@@ -203,6 +207,7 @@ enum SWR_FORMAT
     R8G8B8_UNORM_SRGB           = 0x1A8,
     R16G16B16_UINT              = 0x1B0,
     R16G16B16_SINT              = 0x1B1,
+    R32_SFIXED                  = 0x1B2,
     R10G10B10A2_SNORM           = 0x1B3,
     R10G10B10A2_USCALED         = 0x1B4,
     R10G10B10A2_SSCALED         = 0x1B5,
diff --git a/src/gallium/drivers/swr/rasterizer/core/format_traits.h b/src/gallium/drivers/swr/rasterizer/core/format_traits.h
index d55ee1fab6d..59d4e7d7e9f 100644
--- a/src/gallium/drivers/swr/rasterizer/core/format_traits.h
+++ b/src/gallium/drivers/swr/rasterizer/core/format_traits.h
@@ -20,7 +20,7 @@
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 *
-* @file format_traits.h
+* @file gen_format_traits.h
 *
 * @brief Format Traits.  auto-generated file
 *
@@ -200,6 +200,28 @@ template<> struct FormatTraits<R32G32B32A32_USCALED> :
 };
 
 //////////////////////////////////////////////////////////////////////////
+/// FormatTraits<R32G32B32A32_SFIXED> - Format traits specialization for R32G32B32A32_SFIXED
+//////////////////////////////////////////////////////////////////////////
+template<> struct FormatTraits<R32G32B32A32_SFIXED> :
+    ComponentTraits<SWR_TYPE_SFIXED, 32, SWR_TYPE_SFIXED, 32, SWR_TYPE_SFIXED, 32, SWR_TYPE_SFIXED, 32>,
+    FormatSwizzle<0, 1, 2, 3>,
+    Defaults<0, 0, 0, 0x3f800000>
+{
+    static const uint32_t bpp{ 128 };
+    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 Transpose32_32_32_32    TransposeT;
+    typedef Format4<32, 32, 32, 32> FormatT;
+};
+
+//////////////////////////////////////////////////////////////////////////
 /// FormatTraits<R32G32B32_FLOAT> - Format traits specialization for R32G32B32_FLOAT
 //////////////////////////////////////////////////////////////////////////
 template<> struct FormatTraits<R32G32B32_FLOAT> :
@@ -310,6 +332,28 @@ template<> struct FormatTraits<R32G32B32_USCALED> :
 };
 
 //////////////////////////////////////////////////////////////////////////
+/// FormatTraits<R32G32B32_SFIXED> - Format traits specialization for R32G32B32_SFIXED
+//////////////////////////////////////////////////////////////////////////
+template<> struct FormatTraits<R32G32B32_SFIXED> :
+    ComponentTraits<SWR_TYPE_SFIXED, 32, SWR_TYPE_SFIXED, 32, SWR_TYPE_SFIXED, 32>,
+    FormatSwizzle<0, 1, 2>,
+    Defaults<0, 0, 0, 0x3f800000>
+{
+    static const uint32_t bpp{ 96 };
+    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 Transpose32_32_32   TransposeT;
+    typedef Format3<32, 32, 32> FormatT;
+};
+
+//////////////////////////////////////////////////////////////////////////
 /// FormatTraits<R16G16B16A16_UNORM> - Format traits specialization for R16G16B16A16_UNORM
 //////////////////////////////////////////////////////////////////////////
 template<> struct FormatTraits<R16G16B16A16_UNORM> :
@@ -728,6 +772,28 @@ template<> struct FormatTraits<R32G32_USCALED> :
 };
 
 //////////////////////////////////////////////////////////////////////////
+/// FormatTraits<R32G32_SFIXED> - Format traits specialization for R32G32_SFIXED
+//////////////////////////////////////////////////////////////////////////
+template<> struct FormatTraits<R32G32_SFIXED> :
+    ComponentTraits<SWR_TYPE_SFIXED, 32, SWR_TYPE_SFIXED, 32>,
+    FormatSwizzle<0, 1>,
+    Defaults<0, 0, 0, 0x3f800000>
+{
+    static const uint32_t bpp{ 64 };
+    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 Transpose32_32  TransposeT;
+    typedef Format2<32, 32> FormatT;
+};
+
+//////////////////////////////////////////////////////////////////////////
 /// FormatTraits<B8G8R8A8_UNORM> - Format traits specialization for B8G8R8A8_UNORM
 //////////////////////////////////////////////////////////////////////////
 template<> struct FormatTraits<B8G8R8A8_UNORM> :
@@ -3368,6 +3434,28 @@ template<> struct FormatTraits<R16G16B16_SINT> :
 };
 
 //////////////////////////////////////////////////////////////////////////
+/// FormatTraits<R32_SFIXED> - Format traits specialization for R32_SFIXED
+//////////////////////////////////////////////////////////////////////////
+template<> struct FormatTraits<R32_SFIXED> :
+    ComponentTraits<SWR_TYPE_SFIXED, 32>,
+    FormatSwizzle<0>,
+    Defaults<0, 0, 0, 0x3f800000>
+{
+    static const uint32_t bpp{ 32 };
+    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<32> TransposeT;
+    typedef Format1<32>                  FormatT;
+};
+
+//////////////////////////////////////////////////////////////////////////
 /// FormatTraits<R10G10B10A2_SNORM> - Format traits specialization for R10G10B10A2_SNORM
 //////////////////////////////////////////////////////////////////////////
 template<> struct FormatTraits<R10G10B10A2_SNORM> :
diff --git a/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp b/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp
index bdd818b6b8f..c5936e59bfe 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp
+++ b/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp
@@ -46,6 +46,7 @@ enum ConversionType
     CONVERT_NORMALIZED,
     CONVERT_USCALED,
     CONVERT_SSCALED,
+    CONVERT_SFIXED,
 };
 
 //////////////////////////////////////////////////////////////////////////
@@ -424,6 +425,9 @@ void FetchJit::JitLoadVertices(const FETCH_COMPILE_STATE &fetchState, Value* str
                 case SWR_TYPE_SSCALED:
                     vec = SI_TO_FP(vec, VectorType::get(mFP32Ty, 4));
                     break;
+                case SWR_TYPE_SFIXED:
+                    vec = FMUL(SI_TO_FP(vec, VectorType::get(mFP32Ty, 4)), VBROADCAST(C(1/65536.0f)));
+                    break;
                 case SWR_TYPE_UNKNOWN:
                 case SWR_TYPE_UNUSED:
                     SWR_ASSERT(false, "Unsupported type %d!", info.type[0]);
@@ -943,6 +947,10 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
                     conversionType = CONVERT_SSCALED;
                     extendCastType = Instruction::CastOps::SIToFP;
                     break;
+                case SWR_TYPE_SFIXED:
+                    conversionType = CONVERT_SFIXED;
+                    extendCastType = Instruction::CastOps::SExt;
+                    break;
                 default:
                     break;
             }
@@ -1034,6 +1042,10 @@ void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState,
                                 {
                                     pGather = SI_TO_FP(pGather, mSimdFP32Ty);
                                 }
+                                else if (conversionType == CONVERT_SFIXED)
+                                {
+                                    pGather = FMUL(SI_TO_FP(pGather, mSimdFP32Ty), VBROADCAST(C(1/65536.0f)));
+                                }
 
                                 vVertexElements[currentVertexElement++] = pGather;
                                 // e.g. result of a single 8x32bit integer gather for 32bit components