AltiVec SIMD implementation of H2V1 and H2V2 plain upsampling (used only when decompressing YCCK images with fast upsampling enabled.)

git-svn-id: svn://svn.code.sf.net/p/libjpeg-turbo/code/trunk@1504 632fc199-4ca6-4c93-a231-07263d6284db

4 files changed, 131 insertions, 0 deletions

diff --git a/ChangeLog.txt b/ChangeLog.txt
index 8866455..b61ab2b 100644
--- a/ChangeLog.txt
+++ b/ChangeLog.txt
@@ -1,3 +1,14 @@
+1.5 pre-beta
+============
+
+[1] Added full SIMD acceleration for PowerPC platforms using AltiVec VMX
+(128-bit SIMD) instructions.  Although the performance of libjpeg-turbo on
+PowerPC was already good, due to the increased number of registers available
+to the compiler vs. x86, it was still possible to speed up compression by about
+3-4x and decompression by about 2-2.5x (relative to libjpeg v6b) through the
+use of AltiVec instructions.
+
+
 1.4.0
 =====
 
diff --git a/simd/jdsample-altivec.c b/simd/jdsample-altivec.c
index f49e403..f73b185 100644
--- a/simd/jdsample-altivec.c
+++ b/simd/jdsample-altivec.c
@@ -294,3 +294,91 @@ jsimd_h2v2_fancy_upsample_altivec (int max_v_samp_factor,
     }
   }
 }
+
+
+/* These are rarely used (mainly just for decompressing YCCK images) */
+
+void
+jsimd_h2v1_upsample_altivec (int max_v_samp_factor,
+                             JDIMENSION output_width,
+                             JSAMPARRAY input_data,
+                             JSAMPARRAY * output_data_ptr)
+{
+  JSAMPARRAY output_data = *output_data_ptr;
+  JSAMPROW inptr, outptr;
+  int inrow, incol;
+
+  __vector unsigned char in, inl, inh;
+
+  for (inrow = 0; inrow < max_v_samp_factor; inrow++) {
+    inptr = input_data[inrow];
+    outptr = output_data[inrow];
+
+    for (incol = (output_width + 31) & (~31); incol > 0;
+         incol -= 64, inptr += 32, outptr += 64) {
+
+      in = vec_ld(0, inptr);
+      inl = vec_mergeh(in, in);
+      inh = vec_mergel(in, in);
+
+      vec_st(inl, 0, outptr);
+      vec_st(inh, 16, outptr);
+
+      if (incol > 32) {
+        in = vec_ld(16, inptr);
+        inl = vec_mergeh(in, in);
+        inh = vec_mergel(in, in);
+
+        vec_st(inl, 32, outptr);
+        vec_st(inh, 48, outptr);
+      }
+    }
+  }
+}
+
+
+void
+jsimd_h2v2_upsample_altivec (int max_v_samp_factor,
+                             JDIMENSION output_width,
+                             JSAMPARRAY input_data,
+                             JSAMPARRAY * output_data_ptr)
+{
+  JSAMPARRAY output_data = *output_data_ptr;
+  JSAMPROW inptr, outptr0, outptr1;
+  int inrow, outrow, incol;
+
+  __vector unsigned char in, inl, inh;
+
+  for (inrow = 0, outrow = 0; outrow < max_v_samp_factor; inrow++) {
+
+    inptr = input_data[inrow];
+    outptr0 = output_data[outrow++];
+    outptr1 = output_data[outrow++];
+
+    for (incol = (output_width + 31) & (~31); incol > 0;
+         incol -= 64, inptr += 32, outptr0 += 64, outptr1 += 64) {
+
+      in = vec_ld(0, inptr);
+      inl = vec_mergeh(in, in);
+      inh = vec_mergel(in, in);
+
+      vec_st(inl, 0, outptr0);
+      vec_st(inl, 0, outptr1);
+
+      vec_st(inh, 16, outptr0);
+      vec_st(inh, 16, outptr1);
+
+      if (incol > 32) {
+        in = vec_ld(16, inptr);
+        inl = vec_mergeh(in, in);
+        inh = vec_mergel(in, in);
+
+        vec_st(inl, 32, outptr0);
+        vec_st(inl, 32, outptr1);
+
+        vec_st(inh, 48, outptr0);
+        vec_st(inh, 48, outptr1);
+      }
+    }
+  }
+}
diff --git a/simd/jsimd.h b/simd/jsimd.h
index 4e6e8d7..c0de7e7 100644
--- a/simd/jsimd.h
+++ b/simd/jsimd.h
@@ -425,6 +425,12 @@ EXTERN(void) jsimd_int_upsample_mips_dspr2
          JSAMPARRAY * output_data_ptr, JDIMENSION output_width,
          int max_v_samp_factor);
 
+EXTERN(void) jsimd_h2v1_upsample_altivec
+        (int max_v_samp_factor, JDIMENSION output_width, JSAMPARRAY input_data,
+         JSAMPARRAY * output_data_ptr);
+EXTERN(void) jsimd_h2v2_upsample_altivec
+        (int max_v_samp_factor, JDIMENSION output_width, JSAMPARRAY input_data,
+         JSAMPARRAY * output_data_ptr);
 
 /* Fancy Upsampling */
 EXTERN(void) jsimd_h2v1_fancy_upsample_mmx
diff --git a/simd/jsimd_powerpc.c b/simd/jsimd_powerpc.c
index 1da406e..2fc6814 100644
--- a/simd/jsimd_powerpc.c
+++ b/simd/jsimd_powerpc.c
@@ -280,12 +280,34 @@ jsimd_h2v1_downsample (j_compress_ptr cinfo, jpeg_component_info * compptr,
 GLOBAL(int)
 jsimd_can_h2v2_upsample (void)
 {
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+
+  if (simd_support & JSIMD_ALTIVEC)
+    return 1;
+
   return 0;
 }
 
 GLOBAL(int)
 jsimd_can_h2v1_upsample (void)
 {
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+
+  if (simd_support & JSIMD_ALTIVEC)
+    return 1;
+
   return 0;
 }
 
@@ -295,6 +317,8 @@ jsimd_h2v2_upsample (j_decompress_ptr cinfo,
                      JSAMPARRAY input_data,
                      JSAMPARRAY * output_data_ptr)
 {
+  jsimd_h2v2_upsample_altivec(cinfo->max_v_samp_factor, cinfo->output_width,
+                              input_data, output_data_ptr);
 }
 
 GLOBAL(void)
@@ -303,6 +327,8 @@ jsimd_h2v1_upsample (j_decompress_ptr cinfo,
                      JSAMPARRAY input_data,
                      JSAMPARRAY * output_data_ptr)
 {
+  jsimd_h2v1_upsample_altivec(cinfo->max_v_samp_factor, cinfo->output_width,
+                              input_data, output_data_ptr);
 }
 
 GLOBAL(int)