Merge pull request #153 from tszumski/optimization-changes-v3

Adds avx2 optimization for block encoding and decoding.
aous72 · Sep 9, 2024 · 58ed06d · 58ed06d
2 parents 987c654 + cc8cec4
commit 58ed06d
Show file tree

Hide file tree

Showing 7 changed files with 3,284 additions and 2 deletions.
diff --git a/src/core/CMakeLists.txt b/src/core/CMakeLists.txt
@@ -10,6 +10,7 @@ file(GLOB CODESTREAM_WASM  "codestream/*_wasm.cpp")
 file(GLOB CODING           "coding/*.cpp" "coding/*.h")
 file(GLOB CODING_SSSE3     "coding/*_ssse3.cpp")
 file(GLOB CODING_WASM      "coding/*_wasm.cpp")
+file(GLOB CODING_AVX2      "coding/*_avx2.cpp")
 file(GLOB CODING_AVX512    "coding/*_avx512.cpp")
 file(GLOB COMMON           "common/*.h")
 file(GLOB OTHERS           "others/*.cpp")
@@ -22,7 +23,7 @@ file(GLOB TRANSFORM_AVX512 "transform/*_avx512.cpp")
 file(GLOB TRANSFORM_WASM   "transform/*_wasm.cpp")
 
 list(REMOVE_ITEM CODESTREAM ${CODESTREAM_SSE} ${CODESTREAM_SSE2} ${CODESTREAM_AVX} ${CODESTREAM_AVX2} ${CODESTREAM_WASM})
-list(REMOVE_ITEM CODING ${CODING_SSSE3} ${CODING_WASM} ${CODING_AVX512})
+list(REMOVE_ITEM CODING ${CODING_SSSE3} ${CODING_WASM} ${CODING_AVX2} ${CODING_AVX512})
 list(REMOVE_ITEM TRANSFORM ${TRANSFORM_SSE} ${TRANSFORM_SSE2} ${TRANSFORM_AVX} ${TRANSFORM_AVX2} ${TRANSFORM_AVX512} ${TRANSFORM_WASM})
 list(APPEND SOURCES ${CODESTREAM} ${CODING} ${COMMON} ${OTHERS} ${TRANSFORM})
 
@@ -70,9 +71,10 @@ else()
         source_group("transform" FILES ${TRANSFORM_AVX})
       endif()
       if (NOT OJPH_DISABLE_AVX2)
-        list(APPEND SOURCES ${CODESTREAM_AVX2} ${TRANSFORM_AVX2})
+        list(APPEND SOURCES ${CODESTREAM_AVX2} ${TRANSFORM_AVX2} ${CODING_AVX2})
         source_group("codestream" FILES ${CODESTREAM_AVX2})
         source_group("transform" FILES ${TRANSFORM_AVX2})
+        source_group("coding" FILES ${CODING_AVX2})
       endif()
       if ((NOT OJPH_DISABLE_AVX512) AND ("${OJPH_TARGET_ARCH}" MATCHES "OJPH_ARCH_X86_64"))
         list(APPEND SOURCES ${CODING_AVX512} ${TRANSFORM_AVX512})
@@ -84,6 +86,8 @@ else()
       if (MSVC)
         set_source_files_properties(codestream/ojph_codestream_avx.cpp PROPERTIES COMPILE_FLAGS "/arch:AVX")
         set_source_files_properties(codestream/ojph_codestream_avx2.cpp PROPERTIES COMPILE_FLAGS "/arch:AVX2")
+        set_source_files_properties(coding/ojph_block_decoder_avx2.cpp PROPERTIES COMPILE_FLAGS "/arch:AVX2")
+        set_source_files_properties(coding/ojph_block_encoder_avx2.cpp PROPERTIES COMPILE_FLAGS "/arch:AVX2")
         set_source_files_properties(coding/ojph_block_encoder_avx512.cpp PROPERTIES COMPILE_FLAGS "/arch:AVX512")
         set_source_files_properties(transform/ojph_colour_avx.cpp PROPERTIES COMPILE_FLAGS "/arch:AVX")
         set_source_files_properties(transform/ojph_colour_avx2.cpp PROPERTIES COMPILE_FLAGS "/arch:AVX2")
@@ -94,6 +98,8 @@ else()
         set_source_files_properties(codestream/ojph_codestream_avx.cpp PROPERTIES COMPILE_FLAGS -mavx)
         set_source_files_properties(codestream/ojph_codestream_avx2.cpp PROPERTIES COMPILE_FLAGS -mavx2)
         set_source_files_properties(coding/ojph_block_decoder_ssse3.cpp PROPERTIES COMPILE_FLAGS -mssse3)
+        set_source_files_properties(coding/ojph_block_decoder_avx2.cpp PROPERTIES COMPILE_FLAGS -mavx2)
+        set_source_files_properties(coding/ojph_block_encoder_avx2.cpp PROPERTIES COMPILE_FLAGS -mavx2)
         set_source_files_properties(coding/ojph_block_encoder_avx512.cpp PROPERTIES COMPILE_FLAGS -mavx512cd)
         set_source_files_properties(transform/ojph_colour_avx.cpp PROPERTIES COMPILE_FLAGS -mavx)
         set_source_files_properties(transform/ojph_colour_avx2.cpp PROPERTIES COMPILE_FLAGS -mavx2)

diff --git a/src/core/codestream/ojph_codeblock_fun.cpp b/src/core/codestream/ojph_codeblock_fun.cpp
@@ -169,6 +169,8 @@ namespace ojph {
             tx_to_cb = avx2_irv_tx_to_cb;
             tx_from_cb = avx2_irv_tx_from_cb;
           }
+          encode_cb = ojph_encode_codeblock_avx2;
+          decode_cb = ojph_decode_codeblock_avx2;
         }
       #endif // !OJPH_DISABLE_AVX2
 

diff --git a/src/core/coding/ojph_block_decoder.h b/src/core/coding/ojph_block_decoder.h
@@ -60,6 +60,12 @@ namespace ojph {
         ui32 missing_msbs, ui32 num_passes, ui32 lengths1, ui32 lengths2,
         ui32 width, ui32 height, ui32 stride, bool stripe_causal);
 
+    // AVX2-accelerated decoder
+    bool
+      ojph_decode_codeblock_avx2(ui8* coded_data, ui32* decoded_data,
+        ui32 missing_msbs, ui32 num_passes, ui32 lengths1, ui32 lengths2,
+        ui32 width, ui32 height, ui32 stride, bool stripe_causal);
+
     // WASM SIMD-accelerated decoder
     bool
       ojph_decode_codeblock_wasm(ui8* coded_data, ui32* decoded_data,