swr/rast: Permute work for simd16

Fix slow permutes in PA tri lists under SIMD16 emulation on AVX

Added missing permute (interlane, immediate) to SIMDLIB

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

6 files changed, 33 insertions, 2 deletions

diff --git a/src/gallium/drivers/swr/rasterizer/common/simd16intrin.h b/src/gallium/drivers/swr/rasterizer/common/simd16intrin.h
index 019b26d8cfb..98a8b9b2f9f 100644
--- a/src/gallium/drivers/swr/rasterizer/common/simd16intrin.h
+++ b/src/gallium/drivers/swr/rasterizer/common/simd16intrin.h
@@ -138,6 +138,7 @@ typedef SIMD512                             SIMD16;
 #define _simd16_cmpeq_epi8                  SIMD16::cmpeq_epi8
 #define _simd16_cmpgt_epi8                  SIMD16::cmpgt_epi8
 
+#define _simd16_permute_ps_i(a, i)          SIMD16::permute_ps<i>(a)
 #define _simd16_permute_ps                  SIMD16::permute_ps
 #define _simd16_permute_epi32               SIMD16::permute_epi32
 #define _simd16_sllv_epi32                  SIMD16::sllv_epi32
diff --git a/src/gallium/drivers/swr/rasterizer/common/simdintrin.h b/src/gallium/drivers/swr/rasterizer/common/simdintrin.h
index fce360df9a7..b1471a97250 100644
--- a/src/gallium/drivers/swr/rasterizer/common/simdintrin.h
+++ b/src/gallium/drivers/swr/rasterizer/common/simdintrin.h
@@ -106,6 +106,7 @@ typedef SIMD256                             SIMD;
 #define _simd_cmpgt_epi16                   SIMD::cmpgt_epi16
 #define _simd_cmpeq_epi16                   SIMD::cmpeq_epi16
 #define _simd_movemask_epi8                 SIMD::movemask_epi8
+#define _simd_permute_ps_i(a, i)            SIMD::permute_ps<i>(a)
 #define _simd_permute_ps                    SIMD::permute_ps
 #define _simd_permute_epi32                 SIMD::permute_epi32
 #define _simd_srlv_epi32                    SIMD::srlv_epi32
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 42b45528731..00c094a425a 100644
--- a/src/gallium/drivers/swr/rasterizer/common/simdlib_256_avx.inl
+++ b/src/gallium/drivers/swr/rasterizer/common/simdlib_256_avx.inl
@@ -479,6 +479,12 @@ SIMD_EMU_IWRAPPER_2(packs_epi32);   // See documentation for _mm256_packs_epi32
 SIMD_EMU_IWRAPPER_2(packus_epi16);  // See documentation for _mm256_packus_epi16 and _mm512_packus_epi16
 SIMD_EMU_IWRAPPER_2(packus_epi32);  // See documentation for _mm256_packus_epi32 and _mm512_packus_epi32
 
+template<int ImmT>
+static SIMDINLINE Float SIMDCALL permute_ps(Float const &a)
+{
+    return _mm256_permute_ps(a, ImmT);
+}
+
 static SIMDINLINE Integer SIMDCALL permute_epi32(Integer const &a, Integer const &swiz) // return a[swiz[i]] for each 32-bit lane i (int32)
 {
     Integer result;
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 9cd0a640025..96c24fff9da 100644
--- a/src/gallium/drivers/swr/rasterizer/common/simdlib_256_avx2.inl
+++ b/src/gallium/drivers/swr/rasterizer/common/simdlib_256_avx2.inl
@@ -174,6 +174,13 @@ SIMD_IWRAPPER_2(packs_epi16);   // See documentation for _mm256_packs_epi16 and
 SIMD_IWRAPPER_2(packs_epi32);   // See documentation for _mm256_packs_epi32 and _mm512_packs_epi32
 SIMD_IWRAPPER_2(packus_epi16);  // See documentation for _mm256_packus_epi16 and _mm512_packus_epi16
 SIMD_IWRAPPER_2(packus_epi32);  // See documentation for _mm256_packus_epi32 and _mm512_packus_epi32
+
+template<int ImmT>
+static SIMDINLINE Float SIMDCALL permute_ps(Float const &a)
+{
+    return _mm256_permute_ps(a, ImmT);
+}
+
 SIMD_IWRAPPER_2_(permute_epi32, permutevar8x32_epi32);
 
 static SIMDINLINE Float SIMDCALL permute_ps(Float const &a, Integer const &swiz)    // return a[swiz[i]] for each 32-bit lane i (float)
diff --git a/src/gallium/drivers/swr/rasterizer/common/simdlib_512_avx512.inl b/src/gallium/drivers/swr/rasterizer/common/simdlib_512_avx512.inl
index f3a58f9e1cb..dfe19d3c04a 100644
--- a/src/gallium/drivers/swr/rasterizer/common/simdlib_512_avx512.inl
+++ b/src/gallium/drivers/swr/rasterizer/common/simdlib_512_avx512.inl
@@ -433,6 +433,12 @@ static SIMDINLINE Integer SIMDCALL insert_si(Integer a, SIMD256Impl::Integer b)
 // SIMD_IWRAPPER_2(packus_epi16);  // See documentation for _mm512_packus_epi16 and _mm512_packus_epi16
 // SIMD_IWRAPPER_2(packus_epi32);  // See documentation for _mm512_packus_epi32 and _mm512_packus_epi32
 
+template<int ImmT>
+static SIMDINLINE Float SIMDCALL permute_ps(Float const &a)
+{
+    return _mm512_permute_ps(a, ImmT);
+}
+
 static SIMDINLINE Integer SIMDCALL permute_epi32(Integer a, Integer swiz)    // return a[swiz[i]] for each 32-bit lane i (float)
 {
     return _mm512_permutexvar_epi32(swiz, a);
diff --git a/src/gallium/drivers/swr/rasterizer/common/simdlib_512_emu.inl b/src/gallium/drivers/swr/rasterizer/common/simdlib_512_emu.inl
index 44eba0b126b..5d5120af36a 100644
--- a/src/gallium/drivers/swr/rasterizer/common/simdlib_512_emu.inl
+++ b/src/gallium/drivers/swr/rasterizer/common/simdlib_512_emu.inl
@@ -519,6 +519,16 @@ SIMD_IWRAPPER_2(packs_epi32);      // See documentation for _mm256_packs_epi32 a
 SIMD_IWRAPPER_2(packus_epi16);     // See documentation for _mm256_packus_epi16 and _mm512_packus_epi16
 SIMD_IWRAPPER_2(packus_epi32);     // See documentation for _mm256_packus_epi32 and _mm512_packus_epi32
 
+template<int ImmT>
+static SIMDINLINE Float SIMDCALL permute_ps(Float const &a)
+{
+    return Float
+    {
+        SIMD256T::template permute_ps<ImmT>(a.v8[0]),
+        SIMD256T::template permute_ps<ImmT>(a.v8[1]),
+    };
+}
+
 static SIMDINLINE Integer SIMDCALL permute_epi32(Integer const &a, Integer const &swiz) // return a[swiz[i]] for each 32-bit lane i (int32)
 {
     return castps_si(permute_ps(castsi_ps(a), swiz));
@@ -587,10 +597,10 @@ template <int shuf>
 static SIMDINLINE Integer SIMDCALL permute2f128_si(Integer const &a, Integer const &b)
 {
     return Integer
-	{
+    {
         SIMD256T::template permute2f128_si<((shuf & 0x03) << 0) | ((shuf & 0x0C) << 2)>(a.v8[0], a.v8[1]),
         SIMD256T::template permute2f128_si<((shuf & 0x30) >> 4) | ((shuf & 0xC0) >> 2)>(b.v8[0], b.v8[1]),
-	};
+    };
 }
 
 SIMD_IWRAPPER_2I_1(shuffle_epi32);