# HG changeset patch
# User Paper <paper@tflc.us>
# Date 1729702285 14400
# Node ID 1f070512497f6656fdd45d25eafd4773da1f5b90
# Parent  75ab77f874e284d7259122e43bab5f7e696af983
impl/sse2: unify multiplication implementations into macros

they were basically the same thing anyway

diff -r 75ab77f874e2 -r 1f070512497f include/vec/impl/sse2.h
--- a/include/vec/impl/sse2.h	Wed Oct 23 10:13:25 2024 -0400
+++ b/include/vec/impl/sse2.h	Wed Oct 23 12:51:25 2024 -0400
@@ -26,6 +26,53 @@
 
 #define VEC_SSE2_ALIGNMENT 16
 
+#define VEC_SSE2_MUL_8x16(sign) \
+	VEC_DECL_MUL(sign, 8, 16) \
+	{ \
+		/* unpack and multiply */ \
+		__m128i dst_even = _mm_mullo_epi16(vec1, vec2); \
+		__m128i dst_odd = _mm_mullo_epi16(_mm_srli_epi16(vec1, 8), _mm_srli_epi16(vec2, 8)); \
+	\
+		/* repack */ \
+		return _mm_or_si128( \
+			_mm_slli_epi16(dst_odd, 8), \
+			_mm_srli_epi16(_mm_slli_epi16(dst_even, 8), 8) \
+		); \
+	}
+
+#define VEC_SSE2_MUL_16x8(sign) \
+	VEC_DECL_MUL(sign, 16, 8) \
+	{ \
+		/* we have a real instruction for this */ \
+		return _mm_mullo_epi16(vec1, vec2); \
+	}
+
+#define VEC_SSE2_MUL_32x4(sign) \
+	VEC_DECL_MUL(sign, 32, 4) \
+	{ \
+		/* this was stolen from... somewhere :) */ \
+		__m128i a13    = _mm_shuffle_epi32(vec1, 0xF5);     /* (-,a3,-,a1) */ \
+		__m128i b13    = _mm_shuffle_epi32(vec2, 0xF5);     /* (-,b3,-,b1) */ \
+		__m128i prod02 = _mm_mul_epu32(vec1, vec2);         /* (-,a2*b2,-,a0*b0) */ \
+		__m128i prod13 = _mm_mul_epu32(a13, b13);           /* (-,a3*b3,-,a1*b1) */ \
+		__m128i prod01 = _mm_unpacklo_epi32(prod02,prod13); /* (-,-,a1*b1,a0*b0) */ \
+		__m128i prod23 = _mm_unpackhi_epi32(prod02,prod13); /* (-,-,a3*b3,a2*b2) */ \
+		return _mm_unpacklo_epi64(prod01, prod23);          /* (ab3,ab2,ab1,ab0) */ \
+	}
+
+#define VEC_SSE2_MUL_64x2(sign) \
+	VEC_DECL_MUL(sign, 64, 2) \
+	{ \
+		__m128i ac = _mm_mul_epu32(vec1, vec2); /* ac = (vec1 & UINT32_MAX) * (vec2 & UINT32_MAX); */ \
+		__m128i b  = _mm_srli_epi64(vec1, 32);  /* b = vec1 >> 32; */ \
+		__m128i bc = _mm_mul_epu32(b, vec2);    /* bc = b * (vec2 & UINT32_MAX); */ \
+		__m128i d  = _mm_srli_epi64(vec2, 32);  /* d = vec2 >> 32; */ \
+		__m128i ad = _mm_mul_epu32(vec1, d);    /* ad = (vec1 & UINT32_MAX) * d; */ \
+		__m128i hi = _mm_add_epi64(bc, ad);     /* hi = bc + ad; */ \
+		hi = _mm_slli_epi64(hi, 32);            /* hi <<= 32; */ \
+		return _mm_add_epi64(hi, ac);           /* return ac + hi; */ \
+	}
+
 #define VEC_DEFINE_OPERATIONS(sign, csign, bits, size) \
 	VEC_DECL_LOAD_ALIGNED(sign, bits, size) \
 	{ \
@@ -72,10 +119,11 @@
 		return _mm_xor_si128(vec1, vec2); \
 	} \
 	\
+	VEC_SSE2_MUL_##bits##x##size(sign) \
+	\
 	VEC_GENERIC_SPLAT(sign, csign, bits, size) \
 	VEC_GENERIC_DIVIDE(sign, csign, bits, size) \
 	VEC_GENERIC_SHIFTS(sign, csign, bits, size) \
-	VEC_DECL_MUL(sign, bits, size); \
 	VEC_GENERIC_AVG(sign, bits, size)
 
 #define VEC_DEFINE_COMPARISONS_SIGNED(bits, size) \
@@ -101,15 +149,6 @@
 # define VUINT8x16_ALIGNMENT VEC_SSE2_ALIGNMENT
 VEC_DEFINE_OPERATIONS(u, U, 8, 16)
 VEC_GENERIC_COMPARISONS(u, U, 8, 16)
-VEC_DECL_MUL(u, 8, 16)
-{
-	// unpack and multiply
-	__m128i dst_even = _mm_mullo_epi16(vec1, vec2);
-	__m128i dst_odd = _mm_mullo_epi16(_mm_srli_epi16(vec1, 8), _mm_srli_epi16(vec2, 8));
-
-	// repack
-	return _mm_or_si128(_mm_slli_epi16(dst_odd, 8), _mm_srli_epi16(_mm_slli_epi16(dst_even, 8), 8));
-}
 #endif
 
 #ifndef VEC_VUINT16X8
@@ -120,10 +159,6 @@
 # define VUINT16x8_ALIGNMENT VEC_SSE2_ALIGNMENT
 VEC_DEFINE_OPERATIONS(u, U, 16, 8)
 VEC_GENERIC_COMPARISONS(u, U, 16, 8)
-VEC_DECL_MUL(u, 16, 8)
-{
-	return _mm_mullo_epi16(vec1, vec2);
-}
 #endif
 
 #ifndef VEC_VUINT32X4
@@ -134,17 +169,6 @@
 # define VUINT32x4_ALIGNMENT VEC_SSE2_ALIGNMENT
 VEC_DEFINE_OPERATIONS(u, U, 32, 4)
 VEC_GENERIC_COMPARISONS(u, U, 32, 4)
-VEC_DECL_MUL(u, 32, 4)
-{
-	/* this was stolen from... somewhere :) */
-	__m128i a13    = _mm_shuffle_epi32(vec1, 0xF5);     // (-,a3,-,a1)
-	__m128i b13    = _mm_shuffle_epi32(vec2, 0xF5);     // (-,b3,-,b1)
-	__m128i prod02 = _mm_mul_epu32(vec1, vec2);         // (-,a2*b2,-,a0*b0)
-	__m128i prod13 = _mm_mul_epu32(a13, b13);           // (-,a3*b3,-,a1*b1)
-	__m128i prod01 = _mm_unpacklo_epi32(prod02,prod13); // (-,-,a1*b1,a0*b0) 
-	__m128i prod23 = _mm_unpackhi_epi32(prod02,prod13); // (-,-,a3*b3,a2*b2)
-	return _mm_unpacklo_epi64(prod01, prod23);          // (ab3,ab2,ab1,ab0)
-}
 #endif
 
 #ifndef VEC_VUINT64X2
@@ -157,32 +181,6 @@
 # define VUINT64x2_ALIGNMENT VEC_SSE2_ALIGNMENT
 VEC_DEFINE_OPERATIONS(u, U, 64, 2)
 VEC_GENERIC_COMPARISONS(u, U, 64, 2)
-VEC_DECL_MUL(u, 64, 2)
-{
-	/* ac = (vec1 & 0xFFFFFFFF) * (vec2 & 0xFFFFFFFF); */
-	__m128i ac = _mm_mul_epu32(vec1, vec2);
-
-	/* b = vec1 >> 32; */
-	__m128i b = _mm_srli_epi64(vec1, 32);
-
-	/* bc = b * (vec2 & 0xFFFFFFFF); */
-	__m128i bc = _mm_mul_epu32(b, vec2);
-
-	/* d = vec2 >> 32; */
-	__m128i d = _mm_srli_epi64(vec2, 32);
-
-	/* ad = (vec1 & 0xFFFFFFFF) * d; */
-	__m128i ad = _mm_mul_epu32(vec1, d);
-
-	/* high = bc + ad; */
-	__m128i high = _mm_add_epi64(bc, ad);
-
-	/* high <<= 32; */
-	high = _mm_slli_epi64(high, 32);
-
-	/* return ac + high; */
-	return _mm_add_epi64(high, ac);
-}
 #endif
 
 #ifndef VEC_VINT8X16
@@ -193,15 +191,6 @@
 # define VINT8x16_ALIGNMENT VEC_SSE2_ALIGNMENT
 VEC_DEFINE_OPERATIONS(, , 8, 16)
 VEC_DEFINE_COMPARISONS_SIGNED(8, 16)
-VEC_DECL_MUL(, 8, 16)
-{
-	// unpack and multiply
-	__m128i dst_even = _mm_mullo_epi16(vec1, vec2);
-	__m128i dst_odd = _mm_mullo_epi16(_mm_srli_epi16(vec1, 8), _mm_srli_epi16(vec2, 8));
-
-	// repack
-	return _mm_or_si128(_mm_slli_epi16(dst_odd, 8), _mm_srli_epi16(_mm_slli_epi16(dst_even, 8), 8));
-}
 #endif
 
 #ifndef VEC_VINT16X8
@@ -212,10 +201,6 @@
 # define VINT16x8_ALIGNMENT VEC_SSE2_ALIGNMENT
 VEC_DEFINE_OPERATIONS(, , 16, 8)
 VEC_DEFINE_COMPARISONS_SIGNED(16, 8)
-VEC_DECL_MUL(, 16, 8)
-{
-	return _mm_mullo_epi16(vec1, vec2);
-}
 #endif
 
 #ifndef VEC_VINT32X4
@@ -226,16 +211,6 @@
 # define VINT32x4_ALIGNMENT VEC_SSE2_ALIGNMENT
 VEC_DEFINE_OPERATIONS(, , 32, 4)
 VEC_DEFINE_COMPARISONS_SIGNED(32, 4)
-VEC_DECL_MUL(, 32, 4)
-{
-	__m128i a13    = _mm_shuffle_epi32(vec1, 0xF5);     // (-,a3,-,a1)
-	__m128i b13    = _mm_shuffle_epi32(vec2, 0xF5);     // (-,b3,-,b1)
-	__m128i prod02 = _mm_mul_epu32(vec1, vec2);         // (-,a2*b2,-,a0*b0)
-	__m128i prod13 = _mm_mul_epu32(a13, b13);           // (-,a3*b3,-,a1*b1)
-	__m128i prod01 = _mm_unpacklo_epi32(prod02,prod13); // (-,-,a1*b1,a0*b0) 
-	__m128i prod23 = _mm_unpackhi_epi32(prod02,prod13); // (-,-,a3*b3,a2*b2) 
-	return _mm_unpacklo_epi64(prod01, prod23);          // (ab3,ab2,ab1,ab0)
-}
 #endif
 
 #ifndef VEC_VINT64X2
@@ -248,34 +223,13 @@
 # define VINT64x2_ALIGNMENT VEC_SSE2_ALIGNMENT
 VEC_DEFINE_OPERATIONS(, , 64, 2)
 VEC_GENERIC_COMPARISONS(, , 64, 2)
-VEC_DECL_MUL(, 64, 2)
-{
-	/* ac = (vec1 & 0xFFFFFFFF) * (vec2 & 0xFFFFFFFF); */
-	__m128i ac = _mm_mul_epu32(vec1, vec2);
-
-	/* b = vec1 >> 32; */
-	__m128i b = _mm_srli_epi64(vec1, 32);
-
-	/* bc = b * (vec2 & 0xFFFFFFFF); */
-	__m128i bc = _mm_mul_epu32(b, vec2);
-
-	/* d = vec2 >> 32; */
-	__m128i d = _mm_srli_epi64(vec2, 32);
-
-	/* ad = (vec1 & 0xFFFFFFFF) * d; */
-	__m128i ad = _mm_mul_epu32(vec1, d);
-
-	/* high = bc + ad; */
-	__m128i high = _mm_add_epi64(bc, ad);
-
-	/* high <<= 32; */
-	high = _mm_slli_epi64(high, 32);
-
-	/* return ac + high; */
-	return _mm_add_epi64(high, ac);
-}
 #endif
 
 #undef VEC_DEFINE_OPERATIONS
-#undef VEC_SSE2_8x16_SHIFT
-#undef VEC_SSE2_16x8_SHIFT
+#undef VEC_DEFINE_COMPARISONS_SIGNED
+
+/* multiply */
+#undef VEC_SSE2_MUL_8x16
+#undef VEC_SSE2_MUL_16x8
+#undef VEC_SSE2_MUL_32x4
+#undef VEC_SSE2_MUL_64x2