From 5721da8bf7733fa1596b032b6adc752e5dc93dac Mon Sep 17 00:00:00 2001
From: Ahmet Inan <inan@aicodix.de>
Date: Tue, 13 Feb 2024 13:27:55 +0100
Subject: [PATCH] added vshuf versions for SSE

---
 sse4_1_double.hh    | 22 ++++++++++++++++++++++
 sse4_1_quadruple.hh | 26 ++++++++++++++++++++++++++
 sse4_1_triple.hh    | 24 ++++++++++++++++++++++++
 3 files changed, 72 insertions(+)

diff --git a/sse4_1_double.hh b/sse4_1_double.hh
index 4daa09d..41dc193 100644
--- a/sse4_1_double.hh
+++ b/sse4_1_double.hh
@@ -1272,3 +1272,25 @@ inline SIMD<int32_t, 8> vclamp(SIMD<int32_t, 8> x, int32_t a, int32_t b)
 	return tmp;
 }
 
+template <>
+inline SIMD<uint8_t, 32> vshuf(SIMD<uint8_t, 32> a, SIMD<uint8_t, 32> b)
+{
+	SIMD<uint8_t, 32> tmp;
+	for (int i = 0; i < 2; ++i)
+		tmp.m[i] = _mm_or_si128(
+			_mm_shuffle_epi8(a.m[0], _mm_or_si128(b.m[i], _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(15)))),
+			_mm_shuffle_epi8(a.m[1], _mm_sub_epi8(b.m[i], _mm_set1_epi8(16))));
+	return tmp;
+}
+
+template <>
+inline SIMD<int8_t, 32> vshuf(SIMD<int8_t, 32> a, SIMD<uint8_t, 32> b)
+{
+	SIMD<int8_t, 32> tmp;
+	for (int i = 0; i < 2; ++i)
+		tmp.m[i] = _mm_or_si128(
+			_mm_shuffle_epi8(a.m[0], _mm_or_si128(b.m[i], _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(15)))),
+			_mm_shuffle_epi8(a.m[1], _mm_sub_epi8(b.m[i], _mm_set1_epi8(16))));
+	return tmp;
+}
+
diff --git a/sse4_1_quadruple.hh b/sse4_1_quadruple.hh
index 71d8fa9..3b4db08 100644
--- a/sse4_1_quadruple.hh
+++ b/sse4_1_quadruple.hh
@@ -1272,3 +1272,29 @@ inline SIMD<int32_t, 16> vclamp(SIMD<int32_t, 16> x, int32_t a, int32_t b)
 	return tmp;
 }
 
+template <>
+inline SIMD<uint8_t, 64> vshuf(SIMD<uint8_t, 64> a, SIMD<uint8_t, 64> b)
+{
+	SIMD<uint8_t, 64> tmp;
+	for (int i = 0; i < 4; ++i)
+		tmp.m[i] = _mm_or_si128(_mm_or_si128(_mm_or_si128(
+			_mm_shuffle_epi8(a.m[0], _mm_or_si128(b.m[i], _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(15)))),
+			_mm_shuffle_epi8(a.m[1], _mm_or_si128(_mm_sub_epi8(b.m[i], _mm_set1_epi8(16)), _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(31))))),
+			_mm_shuffle_epi8(a.m[2], _mm_or_si128(_mm_sub_epi8(b.m[i], _mm_set1_epi8(32)), _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(47))))),
+			_mm_shuffle_epi8(a.m[3], _mm_sub_epi8(b.m[i], _mm_set1_epi8(48))));
+	return tmp;
+}
+
+template <>
+inline SIMD<int8_t, 64> vshuf(SIMD<int8_t, 64> a, SIMD<uint8_t, 64> b)
+{
+	SIMD<int8_t, 64> tmp;
+	for (int i = 0; i < 4; ++i)
+		tmp.m[i] = _mm_or_si128(_mm_or_si128(_mm_or_si128(
+			_mm_shuffle_epi8(a.m[0], _mm_or_si128(b.m[i], _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(15)))),
+			_mm_shuffle_epi8(a.m[1], _mm_or_si128(_mm_sub_epi8(b.m[i], _mm_set1_epi8(16)), _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(31))))),
+			_mm_shuffle_epi8(a.m[2], _mm_or_si128(_mm_sub_epi8(b.m[i], _mm_set1_epi8(32)), _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(47))))),
+			_mm_shuffle_epi8(a.m[3], _mm_sub_epi8(b.m[i], _mm_set1_epi8(48))));
+	return tmp;
+}
+
diff --git a/sse4_1_triple.hh b/sse4_1_triple.hh
index 5b5818d..162d16c 100644
--- a/sse4_1_triple.hh
+++ b/sse4_1_triple.hh
@@ -1272,3 +1272,27 @@ inline SIMD<int32_t, 12> vclamp(SIMD<int32_t, 12> x, int32_t a, int32_t b)
 	return tmp;
 }
 
+template <>
+inline SIMD<uint8_t, 48> vshuf(SIMD<uint8_t, 48> a, SIMD<uint8_t, 48> b)
+{
+	SIMD<uint8_t, 48> tmp;
+	for (int i = 0; i < 3; ++i)
+		tmp.m[i] = _mm_or_si128(_mm_or_si128(
+			_mm_shuffle_epi8(a.m[0], _mm_or_si128(b.m[i], _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(15)))),
+			_mm_shuffle_epi8(a.m[1], _mm_or_si128(_mm_sub_epi8(b.m[i], _mm_set1_epi8(16)), _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(31))))),
+			_mm_shuffle_epi8(a.m[2], _mm_sub_epi8(b.m[i], _mm_set1_epi8(32))));
+	return tmp;
+}
+
+template <>
+inline SIMD<int8_t, 48> vshuf(SIMD<int8_t, 48> a, SIMD<uint8_t, 48> b)
+{
+	SIMD<int8_t, 48> tmp;
+	for (int i = 0; i < 3; ++i)
+		tmp.m[i] = _mm_or_si128(_mm_or_si128(
+			_mm_shuffle_epi8(a.m[0], _mm_or_si128(b.m[i], _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(15)))),
+			_mm_shuffle_epi8(a.m[1], _mm_or_si128(_mm_sub_epi8(b.m[i], _mm_set1_epi8(16)), _mm_cmpgt_epi8(b.m[i], _mm_set1_epi8(31))))),
+			_mm_shuffle_epi8(a.m[2], _mm_sub_epi8(b.m[i], _mm_set1_epi8(32))));
+	return tmp;
+}
+