Convert PQ 4 bit code to dynamic dispatch, yy

Summary:
Migration of the 4-bit codecs to dynamic dispatch.

The migration consists in:
- templatizing the SIMD ResultHandlers to the SIMDLevel
- instantiating the AVX2 and AVX512 code in their own files (compile units)
- removing any SIMD dependency from IndexFastScan and IndexIVFFastScan
- adding dispatching code for the SIMD code

Differential Revision: D73581633

Author: mdouze

demos/demo_simd_levels.py

@@ -0,0 +1,54 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+#
+# This source code is licensed under the MIT license found in the
+# LICENSE file in the root directory of this source tree.
+
+import time
+import faiss
+import numpy as np  
+import os
+from collections import defaultdict 
+from faiss.contrib.datasets import SyntheticDataset 
+
+
+print("compile options", faiss.get_compile_options())
+print("SIMD level: ", faiss.SIMDConfig.get_level_name())
+
+
+ds = SyntheticDataset(32, 8000, 10000, 8000)
+
+
+index = faiss.index_factory(ds.d, "PQ16x4fs")
+# index = faiss.index_factory(ds.d, "IVF64,PQ16x4fs")
+# index = faiss.index_factory(ds.d, "SQ8")
+
+index.train(ds.get_train())
+index.add(ds.get_database())
+
+
+if False: 
+    faiss.omp_set_num_threads(1)
+    print("PID=", os.getpid())
+    input("press enter to continue")
+    # for simd_level in faiss.NONE, faiss.AVX2, faiss.AVX512F: 
+    for simd_level in faiss.AVX2, faiss.AVX512F: 
+      
+        faiss.SIMDConfig.set_level(simd_level)
+        print("simd_level=", faiss.SIMDConfig.get_level_name())
+        for run in range(1000): 
+            D, I = index.search(ds.get_queries(), 10)
+
+times = defaultdict(list)
+
+for run in range(10): 
+    for simd_level in faiss.SIMDLevel_NONE, faiss.SIMDLevel_AVX2, faiss.SIMDLevel_AVX512F: 
+        faiss.SIMDConfig.set_level(simd_level)
+
+        t0 = time.time()    
+        D, I = index.search(ds.get_queries(), 10)
+        t1 = time.time()
+
+        times[faiss.SIMDConfig.get_level_name()].append(t1 - t0)
+
+for simd_level in times: 
+    print(f"simd_level={simd_level} search time: {np.mean(times[simd_level])*1000:.3f} ms")

faiss/CMakeLists.txt

@@ -186,11 +186,11 @@ set(FAISS_HEADERS
   impl/residual_quantizer_encode_steps.h
   impl/simd_result_handlers.h
   impl/zerocopy_io.h
-  impl/code_distance/code_distance.h
-  impl/code_distance/code_distance-generic.h
-  impl/code_distance/code_distance-avx2.h
-  impl/code_distance/code_distance-avx512.h
-  impl/code_distance/code_distance-sve.h
+  impl/pq_code_distance/code_distance.h
+  impl/pq_code_distance/code_distance-generic.h
+  impl/pq_code_distance/code_distance-avx2.h
+  impl/pq_code_distance/code_distance-avx512.h
+  impl/pq_code_distance/code_distance-sve.h
   invlists/BlockInvertedLists.h
   invlists/DirectMap.h
   invlists/InvertedLists.h

faiss/IndexAdditiveQuantizerFastScan.cpp

@@ -7,14 +7,10 @@
 
 #include <faiss/IndexAdditiveQuantizerFastScan.h>
 
-#include <cassert>
-#include <memory>
-
 #include <faiss/impl/FaissAssert.h>
 #include <faiss/impl/LocalSearchQuantizer.h>
-#include <faiss/impl/LookupTableScaler.h>
 #include <faiss/impl/ResidualQuantizer.h>
-#include <faiss/impl/pq4_fast_scan.h>
+#include <faiss/impl/pq_4bit/pq4_fast_scan.h>
 #include <faiss/utils/quantize_lut.h>
 #include <faiss/utils/utils.h>
 
@@ -199,12 +195,7 @@ void IndexAdditiveQuantizerFastScan::search(
         return;
     }
 
-    NormTableScaler scaler(norm_scale);
-    if (metric_type == METRIC_L2) {
-        search_dispatch_implem<true>(n, x, k, distances, labels, &scaler);
-    } else {
-        search_dispatch_implem<false>(n, x, k, distances, labels, &scaler);
-    }
+    search_dispatch_implem(n, x, k, distances, labels, norm_scale);
 }
 
 void IndexAdditiveQuantizerFastScan::sa_decode(