swr/rasterizer: improvements in simdlib

1. fix build issues with MSVC 2019 compiler

The MSVC 2019 compiler seems to have an issue with optimized code-gen
when using the _mm256_and_si256() intrinsic.
Only disable use of integer vpand on buggy versions MSVC 2019.
Otherwise allow use of integer vpand intrinsic.

2. Remove unused vec/matrix functionality

Reviewed-by: Alok Hota <[email protected]>

17 files changed, 49 insertions, 492 deletions

diff --git a/src/gallium/drivers/swr/rasterizer/common/simdintrin.h b/src/gallium/drivers/swr/rasterizer/common/simdintrin.h
index df5c3ac6056..ebb4f4b7f11 100644
--- a/src/gallium/drivers/swr/rasterizer/common/simdintrin.h
+++ b/src/gallium/drivers/swr/rasterizer/common/simdintrin.h
@@ -191,57 +191,6 @@ SIMDINLINE SIMD128::Integer _simd_blend4_epi32(SIMD128::Integer const& a, SIMD12
         SIMD128::blend_ps<mask>(SIMD128::castsi_ps(a), SIMD128::castsi_ps(b)));
 }
 
-SIMDINLINE
-void _simd_mov(simdscalar& r, unsigned int rlane, simdscalar& s, unsigned int slane)
-{
-    OSALIGNSIMD(float) rArray[KNOB_SIMD_WIDTH], sArray[KNOB_SIMD_WIDTH];
-    SIMD256::store_ps(rArray, r);
-    SIMD256::store_ps(sArray, s);
-    rArray[rlane] = sArray[slane];
-    r             = SIMD256::load_ps(rArray);
-}
-
-// Populates a simdvector from a vector. So p = xyzw becomes xxxx yyyy zzzz wwww.
-#define _simdvec_load_ps SIMD::vec4_load1_ps
-
-SIMDINLINE
-void _simdvec_mov(simdvector& r, const simdscalar& s)
-{
-    SIMD::vec4_set1_vps(r, s);
-}
-
-SIMDINLINE
-void _simdvec_mov(simdvector& r, const simdvector& v)
-{
-    r = v;
-}
-
-#if 0
-// just move a lane from the source simdvector to dest simdvector
-SIMDINLINE
-void _simdvec_mov(simdvector &r, unsigned int rlane, simdvector& s, unsigned int slane)
-{
-    _simd_mov(r[0], rlane, s[0], slane);
-    _simd_mov(r[1], rlane, s[1], slane);
-    _simd_mov(r[2], rlane, s[2], slane);
-    _simd_mov(r[3], rlane, s[3], slane);
-}
-
-#endif
-
-#define _simdvec_dp3_ps SIMD::vec4_dp3_ps
-#define _simdvec_dp4_ps SIMD::vec4_dp4_ps
-#define _simdvec_rcp_length_ps SIMD::vec4_rcp_length_ps
-#define _simdvec_normalize_ps SIMD::vec4_normalize_ps
-#define _simdvec_mul_ps SIMD::vec4_mul_ps
-#define _simdvec_add_ps SIMD::vec4_add_ps
-#define _simdvec_min_ps SIMD::vec4_min_ps
-#define _simdvec_max_ps SIMD::vec4_max_ps
-#define _simd_mat4x4_vec4_multiply SIMD::mat4x4_vec4_multiply
-#define _simd_mat3x3_vec3_w0_multiply SIMD::mat3x3_vec3_w0_multiply
-#define _simd_mat4x4_vec3_w1_multiply SIMD::mat4x4_vec3_w1_multiply
-#define _simd_mat4x3_vec3_w1_multiply SIMD::mat4x3_vec3_w1_multiply
-
 //////////////////////////////////////////////////////////////////////////
 /// @brief Compute plane equation vA * vX + vB * vY + vC
 SIMDINLINE simdscalar vplaneps(simdscalar const& vA,
diff --git a/src/gallium/drivers/swr/rasterizer/common/simdlib.hpp b/src/gallium/drivers/swr/rasterizer/common/simdlib.hpp
index 153e2af7eae..53793ba101c 100644
--- a/src/gallium/drivers/swr/rasterizer/common/simdlib.hpp
+++ b/src/gallium/drivers/swr/rasterizer/common/simdlib.hpp
@@ -209,339 +209,6 @@ struct SIMDBase : Traits::IsaImpl
     using Integer     = typename Traits::Integer;
     using Vec4        = typename Traits::Vec4;
     using Mask        = typename Traits::Mask;
-
-    static const size_t VECTOR_BYTES = sizeof(Float);
-
-    // Populates a SIMD Vec4 from a non-simd vector. So p = xyzw becomes xxxx yyyy zzzz wwww.
-    static SIMDINLINE void vec4_load1_ps(Vec4& r, const float* p)
-    {
-        r[0] = SIMD::set1_ps(p[0]);
-        r[1] = SIMD::set1_ps(p[1]);
-        r[2] = SIMD::set1_ps(p[2]);
-        r[3] = SIMD::set1_ps(p[3]);
-    }
-
-    static SIMDINLINE void vec4_set1_vps(Vec4& r, Float const& s)
-    {
-        r[0] = s;
-        r[1] = s;
-        r[2] = s;
-        r[3] = s;
-    }
-
-    static SIMDINLINE Float vec4_dp3_ps(const Vec4& v0, const Vec4& v1)
-    {
-        Float tmp, r;
-        r = SIMD::mul_ps(v0[0], v1[0]); // (v0.x*v1.x)
-
-        tmp = SIMD::mul_ps(v0[1], v1[1]); // (v0.y*v1.y)
-        r   = SIMD::add_ps(r, tmp);       // (v0.x*v1.x) + (v0.y*v1.y)
-
-        tmp = SIMD::mul_ps(v0[2], v1[2]); // (v0.z*v1.z)
-        r   = SIMD::add_ps(r, tmp);       // (v0.x*v1.x) + (v0.y*v1.y) + (v0.z*v1.z)
-
-        return r;
-    }
-
-    static SIMDINLINE Float vec4_dp4_ps(const Vec4& v0, const Vec4& v1)
-    {
-        Float tmp, r;
-        r = SIMD::mul_ps(v0[0], v1[0]); // (v0.x*v1.x)
-
-        tmp = SIMD::mul_ps(v0[1], v1[1]); // (v0.y*v1.y)
-        r   = SIMD::add_ps(r, tmp);       // (v0.x*v1.x) + (v0.y*v1.y)
-
-        tmp = SIMD::mul_ps(v0[2], v1[2]); // (v0.z*v1.z)
-        r   = SIMD::add_ps(r, tmp);       // (v0.x*v1.x) + (v0.y*v1.y) + (v0.z*v1.z)
-
-        tmp = SIMD::mul_ps(v0[3], v1[3]); // (v0.w*v1.w)
-        r   = SIMD::add_ps(r, tmp);       // (v0.x*v1.x) + (v0.y*v1.y) + (v0.z*v1.z)
-
-        return r;
-    }
-
-    static SIMDINLINE Float vec4_rcp_length_ps(const Vec4& v)
-    {
-        Float length = vec4_dp4_ps(v, v);
-        return SIMD::rsqrt_ps(length);
-    }
-
-    static SIMDINLINE void vec4_normalize_ps(Vec4& r, const Vec4& v)
-    {
-        Float rcpLength = vec4_rcp_length_ps(v);
-
-        r[0] = SIMD::mul_ps(v[0], rcpLength);
-        r[1] = SIMD::mul_ps(v[1], rcpLength);
-        r[2] = SIMD::mul_ps(v[2], rcpLength);
-        r[3] = SIMD::mul_ps(v[3], rcpLength);
-    }
-
-    static SIMDINLINE void vec4_mul_ps(Vec4& r, const Vec4& v, Float const& s)
-    {
-        r[0] = SIMD::mul_ps(v[0], s);
-        r[1] = SIMD::mul_ps(v[1], s);
-        r[2] = SIMD::mul_ps(v[2], s);
-        r[3] = SIMD::mul_ps(v[3], s);
-    }
-
-    static SIMDINLINE void vec4_mul_ps(Vec4& r, const Vec4& v0, const Vec4& v1)
-    {
-        r[0] = SIMD::mul_ps(v0[0], v1[0]);
-        r[1] = SIMD::mul_ps(v0[1], v1[1]);
-        r[2] = SIMD::mul_ps(v0[2], v1[2]);
-        r[3] = SIMD::mul_ps(v0[3], v1[3]);
-    }
-
-    static SIMDINLINE void vec4_add_ps(Vec4& r, const Vec4& v0, Float const& s)
-    {
-        r[0] = SIMD::add_ps(v0[0], s);
-        r[1] = SIMD::add_ps(v0[1], s);
-        r[2] = SIMD::add_ps(v0[2], s);
-        r[3] = SIMD::add_ps(v0[3], s);
-    }
-
-    static SIMDINLINE void vec4_add_ps(Vec4& r, const Vec4& v0, const Vec4& v1)
-    {
-        r[0] = SIMD::add_ps(v0[0], v1[0]);
-        r[1] = SIMD::add_ps(v0[1], v1[1]);
-        r[2] = SIMD::add_ps(v0[2], v1[2]);
-        r[3] = SIMD::add_ps(v0[3], v1[3]);
-    }
-
-    static SIMDINLINE void vec4_min_ps(Vec4& r, const Vec4& v0, Float const& s)
-    {
-        r[0] = SIMD::min_ps(v0[0], s);
-        r[1] = SIMD::min_ps(v0[1], s);
-        r[2] = SIMD::min_ps(v0[2], s);
-        r[3] = SIMD::min_ps(v0[3], s);
-    }
-
-    static SIMDINLINE void vec4_max_ps(Vec4& r, const Vec4& v0, Float const& s)
-    {
-        r[0] = SIMD::max_ps(v0[0], s);
-        r[1] = SIMD::max_ps(v0[1], s);
-        r[2] = SIMD::max_ps(v0[2], s);
-        r[3] = SIMD::max_ps(v0[3], s);
-    }
-
-    // Matrix4x4 * Vector4
-    //   outVec.x = (m00 * v.x) + (m01 * v.y) + (m02 * v.z) + (m03 * v.w)
-    //   outVec.y = (m10 * v.x) + (m11 * v.y) + (m12 * v.z) + (m13 * v.w)
-    //   outVec.z = (m20 * v.x) + (m21 * v.y) + (m22 * v.z) + (m23 * v.w)
-    //   outVec.w = (m30 * v.x) + (m31 * v.y) + (m32 * v.z) + (m33 * v.w)
-    static SIMDINLINE void SIMDCALL mat4x4_vec4_multiply(Vec4&        result,
-                                                         const float* pMatrix,
-                                                         const Vec4&  v)
-    {
-        Float m;
-        Float r0;
-        Float r1;
-
-        m         = SIMD::load1_ps(pMatrix + 0 * 4 + 0); // m[row][0]
-        r0        = SIMD::mul_ps(m, v[0]);               // (m00 * v.x)
-        m         = SIMD::load1_ps(pMatrix + 0 * 4 + 1); // m[row][1]
-        r1        = SIMD::mul_ps(m, v[1]);               // (m1 * v.y)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y)
-        m         = SIMD::load1_ps(pMatrix + 0 * 4 + 2); // m[row][2]
-        r1        = SIMD::mul_ps(m, v[2]);               // (m2 * v.z)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
-        m         = SIMD::load1_ps(pMatrix + 0 * 4 + 3); // m[row][3]
-        r1        = SIMD::mul_ps(m, v[3]);               // (m3 * v.z)
-        r0        = SIMD::add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * v.w)
-        result[0] = r0;
-
-        m         = SIMD::load1_ps(pMatrix + 1 * 4 + 0); // m[row][0]
-        r0        = SIMD::mul_ps(m, v[0]);               // (m00 * v.x)
-        m         = SIMD::load1_ps(pMatrix + 1 * 4 + 1); // m[row][1]
-        r1        = SIMD::mul_ps(m, v[1]);               // (m1 * v.y)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y)
-        m         = SIMD::load1_ps(pMatrix + 1 * 4 + 2); // m[row][2]
-        r1        = SIMD::mul_ps(m, v[2]);               // (m2 * v.z)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
-        m         = SIMD::load1_ps(pMatrix + 1 * 4 + 3); // m[row][3]
-        r1        = SIMD::mul_ps(m, v[3]);               // (m3 * v.z)
-        r0        = SIMD::add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * v.w)
-        result[1] = r0;
-
-        m         = SIMD::load1_ps(pMatrix + 2 * 4 + 0); // m[row][0]
-        r0        = SIMD::mul_ps(m, v[0]);               // (m00 * v.x)
-        m         = SIMD::load1_ps(pMatrix + 2 * 4 + 1); // m[row][1]
-        r1        = SIMD::mul_ps(m, v[1]);               // (m1 * v.y)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y)
-        m         = SIMD::load1_ps(pMatrix + 2 * 4 + 2); // m[row][2]
-        r1        = SIMD::mul_ps(m, v[2]);               // (m2 * v.z)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
-        m         = SIMD::load1_ps(pMatrix + 2 * 4 + 3); // m[row][3]
-        r1        = SIMD::mul_ps(m, v[3]);               // (m3 * v.z)
-        r0        = SIMD::add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * v.w)
-        result[2] = r0;
-
-        m         = SIMD::load1_ps(pMatrix + 3 * 4 + 0); // m[row][0]
-        r0        = SIMD::mul_ps(m, v[0]);               // (m00 * v.x)
-        m         = SIMD::load1_ps(pMatrix + 3 * 4 + 1); // m[row][1]
-        r1        = SIMD::mul_ps(m, v[1]);               // (m1 * v.y)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y)
-        m         = SIMD::load1_ps(pMatrix + 3 * 4 + 2); // m[row][2]
-        r1        = SIMD::mul_ps(m, v[2]);               // (m2 * v.z)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
-        m         = SIMD::load1_ps(pMatrix + 3 * 4 + 3); // m[row][3]
-        r1        = SIMD::mul_ps(m, v[3]);               // (m3 * v.z)
-        r0        = SIMD::add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * v.w)
-        result[3] = r0;
-    }
-
-    // Matrix4x4 * Vector3 - Direction Vector where w = 0.
-    //   outVec.x = (m00 * v.x) + (m01 * v.y) + (m02 * v.z) + (m03 * 0)
-    //   outVec.y = (m10 * v.x) + (m11 * v.y) + (m12 * v.z) + (m13 * 0)
-    //   outVec.z = (m20 * v.x) + (m21 * v.y) + (m22 * v.z) + (m23 * 0)
-    //   outVec.w = (m30 * v.x) + (m31 * v.y) + (m32 * v.z) + (m33 * 0)
-    static SIMDINLINE void SIMDCALL mat3x3_vec3_w0_multiply(Vec4&        result,
-                                                            const float* pMatrix,
-                                                            const Vec4&  v)
-    {
-        Float m;
-        Float r0;
-        Float r1;
-
-        m         = SIMD::load1_ps(pMatrix + 0 * 4 + 0); // m[row][0]
-        r0        = SIMD::mul_ps(m, v[0]);               // (m00 * v.x)
-        m         = SIMD::load1_ps(pMatrix + 0 * 4 + 1); // m[row][1]
-        r1        = SIMD::mul_ps(m, v[1]);               // (m1 * v.y)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y)
-        m         = SIMD::load1_ps(pMatrix + 0 * 4 + 2); // m[row][2]
-        r1        = SIMD::mul_ps(m, v[2]);               // (m2 * v.z)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
-        result[0] = r0;
-
-        m         = SIMD::load1_ps(pMatrix + 1 * 4 + 0); // m[row][0]
-        r0        = SIMD::mul_ps(m, v[0]);               // (m00 * v.x)
-        m         = SIMD::load1_ps(pMatrix + 1 * 4 + 1); // m[row][1]
-        r1        = SIMD::mul_ps(m, v[1]);               // (m1 * v.y)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y)
-        m         = SIMD::load1_ps(pMatrix + 1 * 4 + 2); // m[row][2]
-        r1        = SIMD::mul_ps(m, v[2]);               // (m2 * v.z)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
-        result[1] = r0;
-
-        m         = SIMD::load1_ps(pMatrix + 2 * 4 + 0); // m[row][0]
-        r0        = SIMD::mul_ps(m, v[0]);               // (m00 * v.x)
-        m         = SIMD::load1_ps(pMatrix + 2 * 4 + 1); // m[row][1]
-        r1        = SIMD::mul_ps(m, v[1]);               // (m1 * v.y)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y)
-        m         = SIMD::load1_ps(pMatrix + 2 * 4 + 2); // m[row][2]
-        r1        = SIMD::mul_ps(m, v[2]);               // (m2 * v.z)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
-        result[2] = r0;
-
-        result[3] = SIMD::setzero_ps();
-    }
-
-    // Matrix4x4 * Vector3 - Position vector where w = 1.
-    //   outVec.x = (m00 * v.x) + (m01 * v.y) + (m02 * v.z) + (m03 * 1)
-    //   outVec.y = (m10 * v.x) + (m11 * v.y) + (m12 * v.z) + (m13 * 1)
-    //   outVec.z = (m20 * v.x) + (m21 * v.y) + (m22 * v.z) + (m23 * 1)
-    //   outVec.w = (m30 * v.x) + (m31 * v.y) + (m32 * v.z) + (m33 * 1)
-    static SIMDINLINE void SIMDCALL mat4x4_vec3_w1_multiply(Vec4&        result,
-                                                            const float* pMatrix,
-                                                            const Vec4&  v)
-    {
-        Float m;
-        Float r0;
-        Float r1;
-
-        m         = SIMD::load1_ps(pMatrix + 0 * 4 + 0); // m[row][0]
-        r0        = SIMD::mul_ps(m, v[0]);               // (m00 * v.x)
-        m         = SIMD::load1_ps(pMatrix + 0 * 4 + 1); // m[row][1]
-        r1        = SIMD::mul_ps(m, v[1]);               // (m1 * v.y)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y)
-        m         = SIMD::load1_ps(pMatrix + 0 * 4 + 2); // m[row][2]
-        r1        = SIMD::mul_ps(m, v[2]);               // (m2 * v.z)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
-        m         = SIMD::load1_ps(pMatrix + 0 * 4 + 3); // m[row][3]
-        r0        = SIMD::add_ps(r0, m); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * 1)
-        result[0] = r0;
-
-        m         = SIMD::load1_ps(pMatrix + 1 * 4 + 0); // m[row][0]
-        r0        = SIMD::mul_ps(m, v[0]);               // (m00 * v.x)
-        m         = SIMD::load1_ps(pMatrix + 1 * 4 + 1); // m[row][1]
-        r1        = SIMD::mul_ps(m, v[1]);               // (m1 * v.y)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y)
-        m         = SIMD::load1_ps(pMatrix + 1 * 4 + 2); // m[row][2]
-        r1        = SIMD::mul_ps(m, v[2]);               // (m2 * v.z)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
-        m         = SIMD::load1_ps(pMatrix + 1 * 4 + 3); // m[row][3]
-        r0        = SIMD::add_ps(r0, m); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * 1)
-        result[1] = r0;
-
-        m         = SIMD::load1_ps(pMatrix + 2 * 4 + 0); // m[row][0]
-        r0        = SIMD::mul_ps(m, v[0]);               // (m00 * v.x)
-        m         = SIMD::load1_ps(pMatrix + 2 * 4 + 1); // m[row][1]
-        r1        = SIMD::mul_ps(m, v[1]);               // (m1 * v.y)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y)
-        m         = SIMD::load1_ps(pMatrix + 2 * 4 + 2); // m[row][2]
-        r1        = SIMD::mul_ps(m, v[2]);               // (m2 * v.z)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
-        m         = SIMD::load1_ps(pMatrix + 2 * 4 + 3); // m[row][3]
-        r0        = SIMD::add_ps(r0, m); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * 1)
-        result[2] = r0;
-
-        m         = SIMD::load1_ps(pMatrix + 3 * 4 + 0); // m[row][0]
-        r0        = SIMD::mul_ps(m, v[0]);               // (m00 * v.x)
-        m         = SIMD::load1_ps(pMatrix + 3 * 4 + 1); // m[row][1]
-        r1        = SIMD::mul_ps(m, v[1]);               // (m1 * v.y)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y)
-        m         = SIMD::load1_ps(pMatrix + 3 * 4 + 2); // m[row][2]
-        r1        = SIMD::mul_ps(m, v[2]);               // (m2 * v.z)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
-        m         = SIMD::load1_ps(pMatrix + 3 * 4 + 3); // m[row][3]
-        result[3] = SIMD::add_ps(r0, m); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * 1)
-    }
-
-    static SIMDINLINE void SIMDCALL mat4x3_vec3_w1_multiply(Vec4&        result,
-                                                            const float* pMatrix,
-                                                            const Vec4&  v)
-    {
-        Float m;
-        Float r0;
-        Float r1;
-
-        m         = SIMD::load1_ps(pMatrix + 0 * 4 + 0); // m[row][0]
-        r0        = SIMD::mul_ps(m, v[0]);               // (m00 * v.x)
-        m         = SIMD::load1_ps(pMatrix + 0 * 4 + 1); // m[row][1]
-        r1        = SIMD::mul_ps(m, v[1]);               // (m1 * v.y)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y)
-        m         = SIMD::load1_ps(pMatrix + 0 * 4 + 2); // m[row][2]
-        r1        = SIMD::mul_ps(m, v[2]);               // (m2 * v.z)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
-        m         = SIMD::load1_ps(pMatrix + 0 * 4 + 3); // m[row][3]
-        r0        = SIMD::add_ps(r0, m); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * 1)
-        result[0] = r0;
-
-        m         = SIMD::load1_ps(pMatrix + 1 * 4 + 0); // m[row][0]
-        r0        = SIMD::mul_ps(m, v[0]);               // (m00 * v.x)
-        m         = SIMD::load1_ps(pMatrix + 1 * 4 + 1); // m[row][1]
-        r1        = SIMD::mul_ps(m, v[1]);               // (m1 * v.y)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y)
-        m         = SIMD::load1_ps(pMatrix + 1 * 4 + 2); // m[row][2]
-        r1        = SIMD::mul_ps(m, v[2]);               // (m2 * v.z)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
-        m         = SIMD::load1_ps(pMatrix + 1 * 4 + 3); // m[row][3]
-        r0        = SIMD::add_ps(r0, m); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * 1)
-        result[1] = r0;
-
-        m         = SIMD::load1_ps(pMatrix + 2 * 4 + 0); // m[row][0]
-        r0        = SIMD::mul_ps(m, v[0]);               // (m00 * v.x)
-        m         = SIMD::load1_ps(pMatrix + 2 * 4 + 1); // m[row][1]
-        r1        = SIMD::mul_ps(m, v[1]);               // (m1 * v.y)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y)
-        m         = SIMD::load1_ps(pMatrix + 2 * 4 + 2); // m[row][2]
-        r1        = SIMD::mul_ps(m, v[2]);               // (m2 * v.z)
-        r0        = SIMD::add_ps(r0, r1);                // (m0 * v.x) + (m1 * v.y) + (m2 * v.z)
-        m         = SIMD::load1_ps(pMatrix + 2 * 4 + 3); // m[row][3]
-        r0        = SIMD::add_ps(r0, m); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * 1)
-        result[2] = r0;
-        result[3] = SIMD::set1_ps(1.0f);
-    }
 }; // struct SIMDBase
 
 using SIMD128 = SIMDBase<SIMDImpl::SIMD128Impl::Traits>;
diff --git a/src/gallium/drivers/swr/rasterizer/common/simdlib_256_avx.inl b/src/gallium/drivers/swr/rasterizer/common/simdlib_256_avx.inl
index 232f43faec7..b5046e48683 100644
--- a/src/gallium/drivers/swr/rasterizer/common/simdlib_256_avx.inl
+++ b/src/gallium/drivers/swr/rasterizer/common/simdlib_256_avx.inl
@@ -222,14 +222,14 @@ SIMD_EMU_IWRAPPER_2(subs_epu8); // return (b > a) ? 0 : (a - b) (uint8)
 //-----------------------------------------------------------------------
 // Logical operations
 //-----------------------------------------------------------------------
-SIMD_WRAPPER_2(and_ps);         // return a & b       (float treated as int)
-SIMD_EMU_IWRAPPER_2(and_si);    // return a & b       (int)
-SIMD_WRAPPER_2(andnot_ps);      // return (~a) & b    (float treated as int)
-SIMD_EMU_IWRAPPER_2(andnot_si); // return (~a) & b    (int)
-SIMD_WRAPPER_2(or_ps);          // return a | b       (float treated as int)
-SIMD_EMU_IWRAPPER_2(or_si);     // return a | b       (int)
-SIMD_WRAPPER_2(xor_ps);         // return a ^ b       (float treated as int)
-SIMD_EMU_IWRAPPER_2(xor_si);    // return a ^ b       (int)
+SIMD_WRAPPER_2(and_ps);                         // return a & b       (float treated as int)
+SIMD_IFWRAPPER_2(and_si, _mm256_and_ps);        // return a & b       (int)
+SIMD_WRAPPER_2(andnot_ps);                      // return (~a) & b    (float treated as int)
+SIMD_IFWRAPPER_2(andnot_si, _mm256_andnot_ps);  // return (~a) & b    (int)
+SIMD_WRAPPER_2(or_ps);                          // return a | b       (float treated as int)
+SIMD_IFWRAPPER_2(or_si, _mm256_or_ps);          // return a | b       (int)
+SIMD_WRAPPER_2(xor_ps);                         // return a ^ b       (float treated as int)
+SIMD_IFWRAPPER_2(xor_si, _mm256_xor_ps);        // return a ^ b       (int)
 
 //-----------------------------------------------------------------------
 // Shift operations
diff --git a/src/gallium/drivers/swr/rasterizer/common/simdlib_256_avx2.inl b/src/gallium/drivers/swr/rasterizer/common/simdlib_256_avx2.inl
index 49650d52442..8fce96dcea4 100644
--- a/src/gallium/drivers/swr/rasterizer/common/simdlib_256_avx2.inl
+++ b/src/gallium/drivers/swr/rasterizer/common/simdlib_256_avx2.inl
@@ -81,6 +81,7 @@
         return _mm256_##op(a, b, ImmT);                                       \
     }
 
+
 //-----------------------------------------------------------------------
 // Floating point arithmetic operations
 //-----------------------------------------------------------------------
@@ -116,7 +117,14 @@ SIMD_IWRAPPER_2(subs_epu8); // return (b > a) ? 0 : (a - b) (uint8)
 //-----------------------------------------------------------------------
 // Logical operations
 //-----------------------------------------------------------------------
-SIMD_IWRAPPER_2_(and_si, and_si256);       // return a & b       (int)
+#if _MSC_VER >= 1920 // && _MSC_FULL_VER < [some_fixed_version]
+// Some versions of MSVC 2019 don't handle constant folding of and_si() correctly.
+// Using and_ps instead inhibits the compiler's constant folding and actually issues
+// the and intrinsic even though both inputs are constant values.
+#else
+// Use native integer and intrinsic
+SIMD_IWRAPPER_2_(and_si, and_si256); // return a & b       (int)
+#endif
 SIMD_IWRAPPER_2_(andnot_si, andnot_si256); // return (~a) & b    (int)
 SIMD_IWRAPPER_2_(or_si, or_si256);         // return a | b       (int)
 SIMD_IWRAPPER_2_(xor_si, xor_si256);       // return a ^ b       (int)
@@ -213,6 +221,10 @@ static SIMDINLINE Float SIMDCALL
     return _mm256_i32gather_ps(p, idx, static_cast<int>(ScaleT));
 }
 
+#if _MSC_VER == 1920 // && _MSC_FULL_VER < [some_fixed_version]
+// Don't use _mm256_mask_i32gather_ps(), the compiler doesn't preserve the mask register
+// correctly in early versions of MSVC 2019
+#else
 // for each element: (mask & (1 << 31)) ? (i32gather_ps<ScaleT>(p, idx), mask = 0) : old
 template <ScaleFactor ScaleT = ScaleFactor::SF_1>
 static SIMDINLINE Float SIMDCALL
@@ -222,6 +234,7 @@ static SIMDINLINE Float SIMDCALL
     // Only for this intrinsic - not sure why. :(
     return _mm256_mask_i32gather_ps(old.v, p, idx.v, mask.v, static_cast<int>(ScaleT));
 }
+#endif
 
 static SIMDINLINE uint32_t SIMDCALL movemask_epi8(Integer const& a)
 {
diff --git a/src/gallium/drivers/swr/rasterizer/common/simdlib_interface.hpp b/src/gallium/drivers/swr/rasterizer/common/simdlib_interface.hpp
index 85c722c92c0..3d31b39ee55 100644
--- a/src/gallium/drivers/swr/rasterizer/common/simdlib_interface.hpp
+++ b/src/gallium/drivers/swr/rasterizer/common/simdlib_interface.hpp
@@ -328,101 +328,5 @@ struct SIMD256 // or SIMD4 or SIMD16
     //=======================================================================
     // Advanced masking interface (currently available only in SIMD16 width)
     //=======================================================================
-
-
-    //=======================================================================
-    // Extended Utility Functions (common to SIMD256 and SIMD16)
-    //=======================================================================
-
-    //-----------------------------------------------------------------------
-    // Extended Types
-    //-----------------------------------------------------------------------
-
-    // Vec4, an SOA SIMD set of 4-dimensional vectors
-    union Vec4
-    {
-        Vec4() = default;
-        Vec4(Float in)
-        {
-            s.x = in;
-            s.y = in;
-            s.z = in;
-            s.w = in;
-        }
-        Vec4(Float x, Float y, Float z, Float w)
-        {
-            s.x = x;
-            s.y = y;
-            s.z = z;
-            s.w = w;
-        }
-
-        Float      v[4];
-        Integer      vi[4];
-        struct
-        {
-            Float  x;
-            Float  y;
-            Float  z;
-            Float  w;
-        } s;
-        Float& operator[] (const int i) { return v[i]; }
-        Float const & operator[] (const int i) const { return v[i]; }
-    };
-
-    //-----------------------------------------------------------------------
-    // Extended Functions
-    //-----------------------------------------------------------------------
-    static void     vec4_set1_ps(Vec4& r, const float *p);                  // r[0] = set1(p[0]), r[1] = set1(p[1]), ...
-    static void     vec4_set1_vps(Vec4& r, Float s);                        // r[0] = s, r[1] = s, ...
-    static Float    vec4_dp3_ps(const Vec4& v0, const Vec4& v1);            // return dp3(v0, v1)
-    static Float    vec4_dp4_ps(const Vec4& v0, const Vec4& v1);            // return dp4(v0, v1)
-    static Float    vec4_rcp_length_ps(const Vec4& v);                      // return 1.0f / sqrt(dp4(v, v))
-    static void     vec4_normalize_ps(Vec4& r, const Vec4& v);              // r = v * rcp_length(v)
-    static void     vec4_mul_ps(Vec4& r, const Vec4& v, Float s);           // r = v * set1_vps(s)
-    static void     vec4_mul_ps(Vec4& r, const Vec4& v0, const Vec4& v1);   // r = v0 * v1
-    static void     vec4_add_ps(Vec4& r, const Vec4& v0, const Vec4& v1);   // r = v0 + v1
-    static void     vec4_min_ps(Vec4& r, const Vec4& v0, Float s);          // r = (v0 < s) ? v0 : s
-    static void     vec4_max_ps(Vec4& r, const Vec4& v0, Float s);          // r = (v0 > s) ? v0 : s
-
-    // Matrix4x4 * Vector4
-    //   result.s.x = (m00 * v.s.x) + (m01 * v.s.y) + (m02 * v.s.z) + (m03 * v.s.w)
-    //   result.s.y = (m10 * v.s.x) + (m11 * v.s.y) + (m12 * v.s.z) + (m13 * v.s.w)
-    //   result.s.z = (m20 * v.s.x) + (m21 * v.s.y) + (m22 * v.s.z) + (m23 * v.s.w)
-    //   result.s.w = (m30 * v.s.x) + (m31 * v.s.y) + (m32 * v.s.z) + (m33 * v.s.w)
-    static void mat4x4_vec4_multiply(
-            Vec4& result,
-            const float *pMatrix,
-            const Vec4& v);
-
-    // Matrix4x4 * Vector3 - Direction Vector where w = 0.
-    //   result.s.x = (m00 * v.s.x) + (m01 * v.s.y) + (m02 * v.s.z) + (m03 * 0)
-    //   result.s.y = (m10 * v.s.x) + (m11 * v.s.y) + (m12 * v.s.z) + (m13 * 0)
-    //   result.s.z = (m20 * v.s.x) + (m21 * v.s.y) + (m22 * v.s.z) + (m23 * 0)
-    //   result.s.w = (m30 * v.s.x) + (m31 * v.s.y) + (m32 * v.s.z) + (m33 * 0)
-    static void mat3x3_vec3_w0_multiply(
-            Vec4& result,
-            const float *pMatrix,
-            const Vec4& v);
-
-    // Matrix4x4 * Vector3 - Position vector where w = 1.
-    //   result.s.x = (m00 * v.s.x) + (m01 * v.s.y) + (m02 * v.s.z) + (m03 * 1)
-    //   result.s.y = (m10 * v.s.x) + (m11 * v.s.y) + (m12 * v.s.z) + (m13 * 1)
-    //   result.s.z = (m20 * v.s.x) + (m21 * v.s.y) + (m22 * v.s.z) + (m23 * 1)
-    //   result.s.w = (m30 * v.s.x) + (m31 * v.s.y) + (m32 * v.s.z) + (m33 * 1)
-    static void mat4x4_vec3_w1_multiply(
-            Vec4& result,
-            const float *pMatrix,
-            const Vec4& v);
-
-    // Matrix4x3 * Vector3 - Position vector where w = 1.
-    //   result.s.x = (m00 * v.s.x) + (m01 * v.s.y) + (m02 * v.s.z) + (m03 * 1)
-    //   result.s.y = (m10 * v.s.x) + (m11 * v.s.y) + (m12 * v.s.z) + (m13 * 1)
-    //   result.s.z = (m20 * v.s.x) + (m21 * v.s.y) + (m22 * v.s.z) + (m23 * 1)
-    //   result.s.w = 1
-    static void mat4x3_vec3_w1_multiply(
-            Vec4& result,
-            const float *pMatrix,
-            const Vec4& v);
 };
 #endif // #if 0
diff --git a/src/gallium/drivers/swr/rasterizer/common/simdlib_types.hpp b/src/gallium/drivers/swr/rasterizer/common/simdlib_types.hpp
index 944c3c23fd3..3ef847d4ca4 100644
--- a/src/gallium/drivers/swr/rasterizer/common/simdlib_types.hpp
+++ b/src/gallium/drivers/swr/rasterizer/common/simdlib_types.hpp
@@ -315,7 +315,7 @@ namespace SIMDImpl
 
     namespace SIMD512Impl
     {
-#if !(defined(__AVX512F__) || defined(_MM_K0_REG))
+#if !(defined(__AVX512F__) || defined(_ZMMINTRIN_H_INCLUDED))
         // Define AVX512 types if not included via immintrin.h.
         // All data members of these types are ONLY to viewed
         // in a debugger.  Do NOT access them via code!
diff --git a/src/gallium/drivers/swr/rasterizer/core/api.cpp b/src/gallium/drivers/swr/rasterizer/core/api.cpp
index 3601aa3f509..f1b0dc03352 100644
--- a/src/gallium/drivers/swr/rasterizer/core/api.cpp
+++ b/src/gallium/drivers/swr/rasterizer/core/api.cpp
@@ -1551,6 +1551,7 @@ void SwrDispatch(HANDLE   hContext,
     pTaskData->threadGroupCountX = threadGroupCountX;
     pTaskData->threadGroupCountY = threadGroupCountY;
     pTaskData->threadGroupCountZ = threadGroupCountZ;
+    pTaskData->enableThreadDispatch = false;
 
     uint32_t totalThreadGroups = threadGroupCountX * threadGroupCountY * threadGroupCountZ;
     uint32_t dcIndex           = pDC->drawId % pContext->MAX_DRAWS_IN_FLIGHT;
diff --git a/src/gallium/drivers/swr/rasterizer/core/api.h b/src/gallium/drivers/swr/rasterizer/core/api.h
index e1ba893296e..8058defb388 100644
--- a/src/gallium/drivers/swr/rasterizer/core/api.h
+++ b/src/gallium/drivers/swr/rasterizer/core/api.h
@@ -588,6 +588,7 @@ SWR_FUNC(void,
          uint32_t threadGroupCountY,
          uint32_t threadGroupCountZ);
 
+
 /// @note this enum needs to be kept in sync with HOTTILE_STATE!
 enum SWR_TILE_STATE
 {
diff --git a/src/gallium/drivers/swr/rasterizer/core/context.h b/src/gallium/drivers/swr/rasterizer/core/context.h
index 5a8656dcfba..8891cc881a3 100644
--- a/src/gallium/drivers/swr/rasterizer/core/context.h
+++ b/src/gallium/drivers/swr/rasterizer/core/context.h
@@ -140,6 +140,7 @@ struct COMPUTE_DESC
     uint32_t threadGroupCountX;
     uint32_t threadGroupCountY;
     uint32_t threadGroupCountZ;
+    bool     enableThreadDispatch;
 };
 
 typedef void (*PFN_WORK_FUNC)(DRAW_CONTEXT* pDC,
diff --git a/src/gallium/drivers/swr/rasterizer/core/pa.h b/src/gallium/drivers/swr/rasterizer/core/pa.h
index 635bf195e4b..c41376ae97b 100644
--- a/src/gallium/drivers/swr/rasterizer/core/pa.h
+++ b/src/gallium/drivers/swr/rasterizer/core/pa.h
@@ -1447,7 +1447,7 @@ struct PA_TESS : PA_STATE
             SIMDSCALARI indices = _simd16_load_si((const SIMDSCALARI*)m_ppIndices[i]);
             if (!m_SOA)
             {
-                indices = _simd16_mul_epi32(indices, _simd16_set1_epi32(vertexStride / 4));
+                indices = _simd16_mullo_epi32(indices, _simd16_set1_epi32(vertexStride / 4));
             }
 #else
             SIMDSCALARI indices = _simd_load_si((const SIMDSCALARI*)m_ppIndices[i]);
diff --git a/src/gallium/drivers/swr/rasterizer/jitter/JitManager.cpp b/src/gallium/drivers/swr/rasterizer/jitter/JitManager.cpp
index fc8dc46d9de..0f78bd661a5 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/JitManager.cpp
+++ b/src/gallium/drivers/swr/rasterizer/jitter/JitManager.cpp
@@ -584,7 +584,7 @@ struct JitCacheFileHeader
     uint64_t GetObjectCRC() const { return m_objCRC; }
 
 private:
-    static const uint64_t JC_MAGIC_NUMBER = 0xfedcba9876543210ULL + 6;
+    static const uint64_t JC_MAGIC_NUMBER = 0xfedcba9876543210ULL + 7;
     static const size_t   JC_STR_MAX_LEN  = 32;
     static const uint32_t JC_PLATFORM_KEY = (LLVM_VERSION_MAJOR << 24) |
                                             (LLVM_VERSION_MINOR << 16) | (LLVM_VERSION_PATCH << 8) |
diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder.cpp b/src/gallium/drivers/swr/rasterizer/jitter/builder.cpp
index 53f11d66db1..30481b43208 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/builder.cpp
+++ b/src/gallium/drivers/swr/rasterizer/jitter/builder.cpp
@@ -61,6 +61,7 @@ namespace SwrJit
         mInt16PtrTy = PointerType::get(mInt16Ty, 0);
         mInt32PtrTy = PointerType::get(mInt32Ty, 0);
         mInt64PtrTy = PointerType::get(mInt64Ty, 0);
+        mHandleTy   = mInt8PtrTy;
 
         mSimd4FP64Ty = VectorType::get(mDoubleTy, 4);
 
diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder.h b/src/gallium/drivers/swr/rasterizer/jitter/builder.h
index 97550fad23d..6e1d94b9e68 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/builder.h
+++ b/src/gallium/drivers/swr/rasterizer/jitter/builder.h
@@ -78,6 +78,7 @@ namespace SwrJit
         // Built in types: scalar
 
         Type* mVoidTy;
+        Type* mHandleTy;
         Type* mInt1Ty;
         Type* mInt8Ty;
         Type* mInt16Ty;
diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder_mem.cpp b/src/gallium/drivers/swr/rasterizer/jitter/builder_mem.cpp
index b183a9e0082..2d8240187c5 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/builder_mem.cpp
+++ b/src/gallium/drivers/swr/rasterizer/jitter/builder_mem.cpp
@@ -237,6 +237,11 @@ namespace SwrJit
         return MASKED_GATHER(pVecSrcPtr, 4, pVecMask, pVecPassthru);
     }
 
+    void Builder::SCATTER_PTR(Value* pVecDstPtr, Value* pVecSrc, Value* pVecMask)
+    {
+        MASKED_SCATTER(pVecSrc, pVecDstPtr, 4, pVecMask);
+    }
+
     void Builder::Gather4(const SWR_FORMAT format,
                           Value*           pSrcBase,
                           Value*           byteOffsets,
diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder_mem.h b/src/gallium/drivers/swr/rasterizer/jitter/builder_mem.h
index 934a8279c2f..fe4c5dd38a4 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/builder_mem.h
+++ b/src/gallium/drivers/swr/rasterizer/jitter/builder_mem.h
@@ -148,6 +148,7 @@ void GATHER4DD(const SWR_FORMAT_INFO& info,
 Value* GATHERPD(Value* src, Value* pBase, Value* indices, Value* mask, uint8_t scale = 1);
 
 Value* GATHER_PTR(Value* pVecSrcPtr, Value* pVecMask, Value* pVecPassthru);
+void SCATTER_PTR(Value* pVecDstPtr, Value* pVecSrc, Value* pVecMask);
 
 virtual void SCATTERPS(Value*         pDst,
                        Value*         vSrc,
diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp b/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp
index 5b06de352dc..6687ead02d3 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp
+++ b/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp
@@ -170,6 +170,16 @@ namespace SwrJit
         return ConstantInt::get(IRB()->getInt1Ty(), (pred ? 1 : 0));
     }
 
+    Value* Builder::VIMMED1(uint64_t i)
+    {
+        return ConstantVector::getSplat(mVWidth, cast<ConstantInt>(C(i)));
+    }
+
+    Value* Builder::VIMMED1_16(uint64_t i)
+    {
+        return ConstantVector::getSplat(mVWidth16, cast<ConstantInt>(C(i)));
+    }
+
     Value* Builder::VIMMED1(int i)
     {
         return ConstantVector::getSplat(mVWidth, cast<ConstantInt>(C(i)));
diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.h b/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.h
index 91e2a32f1a1..3987a5f3476 100644
--- a/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.h
+++ b/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.h
@@ -71,6 +71,9 @@ Constant* CInc(uint32_t base, uint32_t count)
 
 Constant* PRED(bool pred);
 
+Value* VIMMED1(uint64_t i);
+Value* VIMMED1_16(uint64_t i);
+
 Value* VIMMED1(int i);
 Value* VIMMED1_16(int i);