From a3a31149281df195e7a7a316f95845b9ef8e1b34 Mon Sep 17 00:00:00 2001
From: Kenneth Heafield <github@kheafield.com>
Date: Sun, 19 Apr 2020 13:30:26 +0100
Subject: Sum16To32 using variadic templates

---
 tile/multiply.inl | 17 +++++------------
 1 file changed, 5 insertions(+), 12 deletions(-)

diff --git a/tile/multiply.inl b/tile/multiply.inl
index 3bec91f..f1344d7 100644
--- a/tile/multiply.inl
+++ b/tile/multiply.inl
@@ -19,16 +19,10 @@ namespace intgemm {
 namespace INTGEMM_ARCH {
 
 // Upcast 16 to 32 if needed.
-template <class T> struct SumTo32Body;
-template <> struct SumTo32Body<int16_t> {
-  template <class Iterator> INTGEMM_TARGET static inline void body(Register *regs) {
-    Register &reg = regs[Iterator::template I<0>()];
-    reg = madd_epi16(reg, set1_epi16<Register>(1));
-  }
-};
-template <> struct SumTo32Body<int32_t> {
-  template <class Iterator> INTGEMM_TARGET static inline void body(Register *) {}
-};
+template <std::size_t... i> INTGEMM_TARGET static inline void Sum16To32(Register *regs, int16_t, index_sequence<i...>) {
+  unordered_unfurl((regs[i] = madd_epi16(regs[i], set1_epi16<Register>(1)))...);
+}
+template <std::size_t... i> INTGEMM_TARGET static inline void Sum16To32(Register *, int32_t, index_sequence<i...>) {}
 
 /* Multiply assuming the matrix sizes are a multiple of the kernel size. */
 template <class AccessT, class Kernel> INTGEMM_TARGET __attribute__((flatten)) static inline void MultiplyNoOverhang(AccessT access, const Tile shape) {
@@ -54,8 +48,7 @@ template <class AccessT, class Kernel> INTGEMM_TARGET __attribute__((flatten)) s
         Kernel::Run(reg_access.AAdd(0, inner).BAdd(inner, 0));
       }
 
-      // If 16-bit, upcast to 32-bit while horizontally adding.
-      StaticLoop<SumTo32Body<typename Kernel::Packed::C>, MakeStaticLoopIterator<Outputs>>(c_regs);
+      Sum16To32(c_regs, typename Kernel::Packed::C(), make_index_sequence<Outputs>());
       // Horizontally add 32-bit values.
       Reduce32<Outputs, Sum32Op>(c_regs);
       col_row.CAccessor().template Write<Kernel::kTile.A_rows, Kernel::kTile.B_cols>(c_regs);
-- 
cgit v1.2.3