[FFmpeg-devel] [PATCH 3/6] avcodec/x86: modify simple_idct10 macros to add an action paramter

[James Darnley]

---
 libavcodec/x86/proresdsp.asm              |  2 +-
 libavcodec/x86/simple_idct10.asm          |  8 +++----
 libavcodec/x86/simple_idct10_template.asm | 37 +++++++++++++++++--------------
 3 files changed, 25 insertions(+), 22 deletions(-)

diff --git a/libavcodec/x86/proresdsp.asm b/libavcodec/x86/proresdsp.asm
index 8318a81c5e..3be0ff7757 100644
--- a/libavcodec/x86/proresdsp.asm
+++ b/libavcodec/x86/proresdsp.asm
@@ -52,7 +52,7 @@ SECTION .text
 
 %macro idct_fn 0
 cglobal prores_idct_put_10, 4, 4, 15, pixels, lsize, block, qmat
-    IDCT_FN    pw_1, 15, pw_88, 18, pw_4, pw_1019, r3
+    IDCT_FN    pw_1, 15, pw_88, 18, "put", pw_4, pw_1019, r3
     RET
 %endmacro
 
diff --git a/libavcodec/x86/simple_idct10.asm b/libavcodec/x86/simple_idct10.asm
index 7cfd33eaa3..1a5a2eae9b 100644
--- a/libavcodec/x86/simple_idct10.asm
+++ b/libavcodec/x86/simple_idct10.asm
@@ -69,24 +69,24 @@ SECTION .text
 
 %macro idct_fn 0
 cglobal simple_idct10, 1, 1, 16, block
-    IDCT_FN    "", 12, "", 19
+    IDCT_FN    "", 12, "", 19, "store"
     RET
 
 cglobal simple_idct10_put, 3, 3, 16, pixels, lsize, block
-    IDCT_FN    "", 12, "", 19, 0, pw_1023
+    IDCT_FN    "", 12, "", 19, "put", 0, pw_1023
     RET
 
 cglobal simple_idct12, 1, 1, 16, block
     ; coeffs are already 15bits, adding the offset would cause
     ; overflow in the input
-    IDCT_FN    "", 15, pw_2, 16
+    IDCT_FN    "", 15, pw_2, 16, "store"
     RET
 
 cglobal simple_idct12_put, 3, 3, 16, pixels, lsize, block
     ; range isn't known, so the C simple_idct range is used
     ; Also, using a bias on input overflows, so use the bias
     ; on output of the first butterfly instead
-    IDCT_FN    "", 15, pw_2, 16, 0, pw_4095
+    IDCT_FN    "", 15, pw_2, 16, "put", 0, pw_4095
     RET
 %endmacro
 
diff --git a/libavcodec/x86/simple_idct10_template.asm b/libavcodec/x86/simple_idct10_template.asm
index 3f398985a5..8367011dfd 100644
--- a/libavcodec/x86/simple_idct10_template.asm
+++ b/libavcodec/x86/simple_idct10_template.asm
@@ -218,11 +218,12 @@
 ; %2 = row bias macro
 ; %3 = column shift
 ; %4 = column bias macro
-; %5 = min pixel value
-; %6 = max pixel value
-; %7 = qmat (for prores)
+; %5 = final action (nothing, "store", "put", "add")
+; %6 = min pixel value
+; %7 = max pixel value
+; %8 = qmat (for prores)
 
-%macro IDCT_FN 4-7
+%macro IDCT_FN 4-8
     ; for (i = 0; i < 8; i++)
     ;     idctRowCondDC(block + i*8);
     mova        m10,[blockq+ 0]        ; { row[0] }[0-7]
@@ -230,13 +231,13 @@
     mova        m13,[blockq+64]        ; { row[4] }[0-7]
     mova        m12,[blockq+96]        ; { row[6] }[0-7]
 
-%if %0 == 7
-    pmullw      m10,[%7+ 0]
-    pmullw      m8, [%7+32]
-    pmullw      m13,[%7+64]
-    pmullw      m12,[%7+96]
+%if %0 == 8
+    pmullw      m10,[%8+ 0]
+    pmullw      m8, [%8+32]
+    pmullw      m13,[%8+64]
+    pmullw      m12,[%8+96]
 
-    IDCT_1D     %1, %2, %7
+    IDCT_1D     %1, %2, %8
 %else
     IDCT_1D     %1, %2
 %endif
@@ -257,7 +258,8 @@
     IDCT_1D     %3, %4
 
     ; clip/store
-%if %0 == 4
+%if %0 >= 5
+%ifidn %5,"store"
     ; No clamping, means pure idct
     mova  [blockq+  0], m8
     mova  [blockq+ 16], m0
@@ -267,13 +269,13 @@
     mova  [blockq+ 80], m11
     mova  [blockq+ 96], m9
     mova  [blockq+112], m10
-%else
-%ifidn %5, 0
+%elifidn %5,"put"
+%ifidn %6, 0
     pxor        m3, m3
 %else
-    mova        m3, [%5]
-%endif
-    mova        m5, [%6]
+    mova        m3, [%6]
+%endif ; ifidn %6, 0
+    mova        m5, [%7]
     pmaxsw      m8,  m3
     pmaxsw      m0,  m3
     pmaxsw      m1,  m3
@@ -301,7 +303,8 @@
     mova  [r0+r1  ], m11
     mova  [r0+r1*2], m9
     mova  [r0+r2  ], m10
-%endif
+%endif ; %5 action
+%endif; if %0 >= 5
 %endmacro
 
 %endif
-- 
2.13.1