Store the operator table more compactly.

The four cases for each operator:

    none-are-opaque, src-is-opaque, dest-is-opaque, both-are-opaque

are packed into one uint32_t per operator. The relevant strength
reduced operator can then be found by packing the source-is-opaque and
dest-is-opaque into two bits and shifting that number of bytes.

Chris Wilson pointed out a bug in the original version of this commit:
dest_is_opaque and source_is_opaque were used as booleans, but their
actual values were the results of a logical AND with the
FAST_PATH_OPAQUE flag, so the shift value was wildly wrong.

The only reason it actually passed the test suite (on x86) was that
the compiler computed the shift amount in the cl register, and the low
byte of FAST_PATH_OPAQUE happens to be 0, so no shifting actually took
place, and the original operator was returned.

1 files changed, 84 insertions, 98 deletions

diff --git a/pixman/pixman.c b/pixman/pixman.c
index bbd978cc..21565298 100644
--- a/pixman/pixman.c
+++ b/pixman/pixman.c
@@ -30,97 +30,90 @@
 
 #include <stdlib.h>
 
-/*
- * Operator optimizations based on source or destination opacity
- */
-typedef struct
-{
-    pixman_op_t op;
-    pixman_op_t op_src_dst_opaque;
-    pixman_op_t op_src_opaque;
-    pixman_op_t op_dst_opaque;
-} operator_info_t;
+static pixman_implementation_t *imp;
 
-#define NO_OPTIMIZATION(a) (a), (a), (a), (a)
+#define PACK(a, b, c, d)			\
+    (((uint8_t)PIXMAN_OP_ ## a <<  0)	|	\
+     ((uint8_t)PIXMAN_OP_ ## b <<  8)	|	\
+     ((uint8_t)PIXMAN_OP_ ## c << 16)	|	\
+     ((uint8_t)PIXMAN_OP_ ## d << 24))
 
-static const operator_info_t operator_table[] =
+static const uint32_t operator_table[] =
 {
-    /* Input Operator           SRC&DST Opaque          SRC Opaque              DST Opaque      */
-    { NO_OPTIMIZATION (PIXMAN_OP_CLEAR) },
-    { NO_OPTIMIZATION (PIXMAN_OP_SRC) },
-    { NO_OPTIMIZATION (PIXMAN_OP_DST) },
-    { PIXMAN_OP_OVER,           PIXMAN_OP_SRC,          PIXMAN_OP_SRC,          PIXMAN_OP_OVER },
-    { PIXMAN_OP_OVER_REVERSE,   PIXMAN_OP_DST,          PIXMAN_OP_OVER_REVERSE, PIXMAN_OP_DST },
-    { PIXMAN_OP_IN,             PIXMAN_OP_SRC,          PIXMAN_OP_IN,           PIXMAN_OP_SRC },
-    { PIXMAN_OP_IN_REVERSE,     PIXMAN_OP_DST,          PIXMAN_OP_DST,          PIXMAN_OP_IN_REVERSE },
-    { PIXMAN_OP_OUT,            PIXMAN_OP_CLEAR,        PIXMAN_OP_OUT,          PIXMAN_OP_CLEAR },
-    { PIXMAN_OP_OUT_REVERSE,    PIXMAN_OP_CLEAR,        PIXMAN_OP_CLEAR,        PIXMAN_OP_OUT_REVERSE },
-    { PIXMAN_OP_ATOP,           PIXMAN_OP_SRC,          PIXMAN_OP_IN,           PIXMAN_OP_OVER },
-    { PIXMAN_OP_ATOP_REVERSE,   PIXMAN_OP_DST,          PIXMAN_OP_OVER_REVERSE, PIXMAN_OP_IN_REVERSE },
-    { PIXMAN_OP_XOR,            PIXMAN_OP_CLEAR,        PIXMAN_OP_OUT,          PIXMAN_OP_OUT_REVERSE },
-    { NO_OPTIMIZATION (PIXMAN_OP_ADD) },
-    { PIXMAN_OP_SATURATE,       PIXMAN_OP_DST,          PIXMAN_OP_OVER_REVERSE, PIXMAN_OP_DST },
-
-    { PIXMAN_OP_NONE /* 0x0e */ },
-    { PIXMAN_OP_NONE /* 0x0f */ },
-
-    { NO_OPTIMIZATION (PIXMAN_OP_DISJOINT_CLEAR) },
-    { NO_OPTIMIZATION (PIXMAN_OP_DISJOINT_SRC) },
-    { NO_OPTIMIZATION (PIXMAN_OP_DISJOINT_DST) },
-    { NO_OPTIMIZATION (PIXMAN_OP_DISJOINT_OVER) },
-    { NO_OPTIMIZATION (PIXMAN_OP_DISJOINT_OVER_REVERSE) },
-    { NO_OPTIMIZATION (PIXMAN_OP_DISJOINT_IN) },
-    { NO_OPTIMIZATION (PIXMAN_OP_DISJOINT_IN_REVERSE) },
-    { NO_OPTIMIZATION (PIXMAN_OP_DISJOINT_OUT) },
-    { NO_OPTIMIZATION (PIXMAN_OP_DISJOINT_OUT_REVERSE) },
-    { NO_OPTIMIZATION (PIXMAN_OP_DISJOINT_ATOP) },
-    { NO_OPTIMIZATION (PIXMAN_OP_DISJOINT_ATOP_REVERSE) },
-    { NO_OPTIMIZATION (PIXMAN_OP_DISJOINT_XOR) },
-
-    { PIXMAN_OP_NONE /* 0x1c */ },
-    { PIXMAN_OP_NONE /* 0x1d */ },
-    { PIXMAN_OP_NONE /* 0x1e */ },
-    { PIXMAN_OP_NONE /* 0x1f */ },
-
-    { NO_OPTIMIZATION (PIXMAN_OP_CONJOINT_CLEAR) },
-    { NO_OPTIMIZATION (PIXMAN_OP_CONJOINT_SRC) },
-    { NO_OPTIMIZATION (PIXMAN_OP_CONJOINT_DST) },
-    { NO_OPTIMIZATION (PIXMAN_OP_CONJOINT_OVER) },
-    { NO_OPTIMIZATION (PIXMAN_OP_CONJOINT_OVER_REVERSE) },
-    { NO_OPTIMIZATION (PIXMAN_OP_CONJOINT_IN) },
-    { NO_OPTIMIZATION (PIXMAN_OP_CONJOINT_IN_REVERSE) },
-    { NO_OPTIMIZATION (PIXMAN_OP_CONJOINT_OUT) },
-    { NO_OPTIMIZATION (PIXMAN_OP_CONJOINT_OUT_REVERSE) },
-    { NO_OPTIMIZATION (PIXMAN_OP_CONJOINT_ATOP) },
-    { NO_OPTIMIZATION (PIXMAN_OP_CONJOINT_ATOP_REVERSE) },
-    { NO_OPTIMIZATION (PIXMAN_OP_CONJOINT_XOR) },
-
-    { PIXMAN_OP_NONE /* 0x2c */ },
-    { PIXMAN_OP_NONE /* 0x2d */ },
-    { PIXMAN_OP_NONE /* 0x2e */ },
-    { PIXMAN_OP_NONE /* 0x2f */ },
-
-    { NO_OPTIMIZATION (PIXMAN_OP_MULTIPLY) },
-    { NO_OPTIMIZATION (PIXMAN_OP_SCREEN) },
-    { NO_OPTIMIZATION (PIXMAN_OP_OVERLAY) },
-    { NO_OPTIMIZATION (PIXMAN_OP_DARKEN) },
-    { NO_OPTIMIZATION (PIXMAN_OP_LIGHTEN) },
-    { NO_OPTIMIZATION (PIXMAN_OP_COLOR_DODGE) },
-    { NO_OPTIMIZATION (PIXMAN_OP_COLOR_BURN) },
-    { NO_OPTIMIZATION (PIXMAN_OP_HARD_LIGHT) },
-    { NO_OPTIMIZATION (PIXMAN_OP_SOFT_LIGHT) },
-    { NO_OPTIMIZATION (PIXMAN_OP_DIFFERENCE) },
-    { NO_OPTIMIZATION (PIXMAN_OP_EXCLUSION) },
-    { NO_OPTIMIZATION (PIXMAN_OP_HSL_HUE) },
-    { NO_OPTIMIZATION (PIXMAN_OP_HSL_SATURATION) },
-    { NO_OPTIMIZATION (PIXMAN_OP_HSL_COLOR) },
-    { NO_OPTIMIZATION (PIXMAN_OP_HSL_LUMINOSITY) },
-
-    { PIXMAN_OP_NONE }
+    /*    Neither Opaque         Src Opaque             Dst Opaque             Both Opaque */
+    PACK (CLEAR,                 CLEAR,                 CLEAR,                 CLEAR),
+    PACK (SRC,                   SRC,                   SRC,                   SRC),
+    PACK (DST,                   DST,                   DST,                   DST),
+    PACK (OVER,                  SRC,                   OVER,                  SRC),
+    PACK (OVER_REVERSE,          OVER_REVERSE,          DST,                   DST),
+    PACK (IN,                    IN,                    SRC,                   SRC),
+    PACK (IN_REVERSE,            DST,                   IN_REVERSE,            DST),
+    PACK (OUT,                   OUT,                   CLEAR,                 CLEAR),
+    PACK (OUT_REVERSE,           CLEAR,                 OUT_REVERSE,           CLEAR),
+    PACK (ATOP,                  IN,                    OVER,                  SRC),
+    PACK (ATOP_REVERSE,          OVER_REVERSE,          IN_REVERSE,            DST),
+    PACK (XOR,                   OUT,                   OUT_REVERSE,           CLEAR),
+    PACK (ADD,                   ADD,                   ADD,                   ADD),
+    PACK (SATURATE,              OVER_REVERSE,          DST,                   DST),
+
+    0 /* 0x0e */,
+    0 /* 0x0f */,
+
+    PACK (DISJOINT_CLEAR,        DISJOINT_CLEAR,        DISJOINT_CLEAR,        DISJOINT_CLEAR),
+    PACK (DISJOINT_SRC,          DISJOINT_SRC,          DISJOINT_SRC,          DISJOINT_SRC),
+    PACK (DISJOINT_DST,          DISJOINT_DST,          DISJOINT_DST,          DISJOINT_DST),
+    PACK (DISJOINT_OVER,         DISJOINT_OVER,         DISJOINT_OVER,         DISJOINT_OVER),
+    PACK (DISJOINT_OVER_REVERSE, DISJOINT_OVER_REVERSE, DISJOINT_OVER_REVERSE, DISJOINT_OVER_REVERSE),
+    PACK (DISJOINT_IN,           DISJOINT_IN,           DISJOINT_IN,           DISJOINT_IN),
+    PACK (DISJOINT_IN_REVERSE,   DISJOINT_IN_REVERSE,   DISJOINT_IN_REVERSE,   DISJOINT_IN_REVERSE),
+    PACK (DISJOINT_OUT,          DISJOINT_OUT,          DISJOINT_OUT,          DISJOINT_OUT),
+    PACK (DISJOINT_OUT_REVERSE,  DISJOINT_OUT_REVERSE,  DISJOINT_OUT_REVERSE,  DISJOINT_OUT_REVERSE),
+    PACK (DISJOINT_ATOP,         DISJOINT_ATOP,         DISJOINT_ATOP,         DISJOINT_ATOP),
+    PACK (DISJOINT_ATOP_REVERSE, DISJOINT_ATOP_REVERSE, DISJOINT_ATOP_REVERSE, DISJOINT_ATOP_REVERSE),
+    PACK (DISJOINT_XOR,          DISJOINT_XOR,          DISJOINT_XOR,          DISJOINT_XOR),
+
+    0 /* 0x1c */,
+    0 /* 0x1d */,
+    0 /* 0x1e */,
+    0 /* 0x1f */,
+
+    PACK (CONJOINT_CLEAR,        CONJOINT_CLEAR,        CONJOINT_CLEAR,        CONJOINT_CLEAR),
+    PACK (CONJOINT_SRC,          CONJOINT_SRC,          CONJOINT_SRC,          CONJOINT_SRC),
+    PACK (CONJOINT_DST,          CONJOINT_DST,          CONJOINT_DST,          CONJOINT_DST),
+    PACK (CONJOINT_OVER,         CONJOINT_OVER,         CONJOINT_OVER,         CONJOINT_OVER),
+    PACK (CONJOINT_OVER_REVERSE, CONJOINT_OVER_REVERSE, CONJOINT_OVER_REVERSE, CONJOINT_OVER_REVERSE),
+    PACK (CONJOINT_IN,           CONJOINT_IN,           CONJOINT_IN,           CONJOINT_IN),
+    PACK (CONJOINT_IN_REVERSE,   CONJOINT_IN_REVERSE,   CONJOINT_IN_REVERSE,   CONJOINT_IN_REVERSE),
+    PACK (CONJOINT_OUT,          CONJOINT_OUT,          CONJOINT_OUT,          CONJOINT_OUT),
+    PACK (CONJOINT_OUT_REVERSE,  CONJOINT_OUT_REVERSE,  CONJOINT_OUT_REVERSE,  CONJOINT_OUT_REVERSE),
+    PACK (CONJOINT_ATOP,         CONJOINT_ATOP,         CONJOINT_ATOP,         CONJOINT_ATOP),
+    PACK (CONJOINT_ATOP_REVERSE, CONJOINT_ATOP_REVERSE, CONJOINT_ATOP_REVERSE, CONJOINT_ATOP_REVERSE),
+    PACK (CONJOINT_XOR,          CONJOINT_XOR,          CONJOINT_XOR,          CONJOINT_XOR),
+
+    0 /* 0x2c */,
+    0 /* 0x2d */,
+    0 /* 0x2e */,
+    0 /* 0x2f */,
+
+    PACK (MULTIPLY,              MULTIPLY,              MULTIPLY,              MULTIPLY),
+    PACK (SCREEN,                SCREEN,                SCREEN,                SCREEN),
+    PACK (OVERLAY,               OVERLAY,               OVERLAY,               OVERLAY),
+    PACK (DARKEN,                DARKEN,                DARKEN,                DARKEN),
+    PACK (LIGHTEN,               LIGHTEN,               LIGHTEN,               LIGHTEN),
+    PACK (COLOR_DODGE,           COLOR_DODGE,           COLOR_DODGE,           COLOR_DODGE),
+    PACK (COLOR_BURN,            COLOR_BURN,            COLOR_BURN,            COLOR_BURN),
+    PACK (HARD_LIGHT,            HARD_LIGHT,            HARD_LIGHT,            HARD_LIGHT),
+    PACK (SOFT_LIGHT,            SOFT_LIGHT,            SOFT_LIGHT,            SOFT_LIGHT),
+    PACK (DIFFERENCE,            DIFFERENCE,            DIFFERENCE,            DIFFERENCE),
+    PACK (EXCLUSION,             EXCLUSION,             EXCLUSION,             EXCLUSION),
+    PACK (HSL_HUE,               HSL_HUE,               HSL_HUE,               HSL_HUE),
+    PACK (HSL_SATURATION,        HSL_SATURATION,        HSL_SATURATION,        HSL_SATURATION),
+    PACK (HSL_COLOR,             HSL_COLOR,             HSL_COLOR,             HSL_COLOR),
+    PACK (HSL_LUMINOSITY,        HSL_LUMINOSITY,        HSL_LUMINOSITY,        HSL_LUMINOSITY),
+
+    0
 };
 
-static pixman_implementation_t *imp;
-
 /*
  * Optimize the current operator based on opacity of source or destination
  * The output operator should be mathematically equivalent to the source.
@@ -131,22 +124,15 @@ optimize_operator (pixman_op_t     op,
 		   uint32_t        mask_flags,
 		   uint32_t        dst_flags)
 {
-    const operator_info_t *info = &(operator_table[op]);
     pixman_bool_t is_source_opaque, is_dest_opaque;
+    int shift;
 
-    assert (info->op == op);
-
-    is_source_opaque = (src_flags & mask_flags) & FAST_PATH_IS_OPAQUE;
-    is_dest_opaque = dst_flags & FAST_PATH_IS_OPAQUE;
+    is_source_opaque = ((src_flags & mask_flags) & FAST_PATH_IS_OPAQUE) != 0;
+    is_dest_opaque = (dst_flags & FAST_PATH_IS_OPAQUE) != 0;
 
-    if (is_source_opaque && is_dest_opaque)
-	return info->op_src_dst_opaque;
-    else if (is_source_opaque)
-	return info->op_src_opaque;
-    else if (is_dest_opaque)
-	return info->op_dst_opaque;
+    shift = 8 * ((is_dest_opaque << 1) | is_source_opaque);
 
-    return op;
+    return (operator_table[op] >> shift) & 0xff;
 }
 
 static void