mesa: Create _mesa_bitcount_64() to replace i965's brw_count_bits()

The i965 driver already had a function to count bits in a 64-bit uint
(brw_count_bits()), but it was buggy (it only counted the bottom 32
bits) and it was clumsy (it had a strange and broken fallback for
non-GCC-like compilers, which fortunately was never used).  Since Mesa
already has a _mesa_bitcount() function, it seems better to just
create a _mesa_bitcount_64() function rather than special-case this in
the i965 driver.

This patch creates the new _mesa_bitcount_64() function and rewrites
all of the old brw_count_bits() calls to refer to it.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

src/mesa/drivers/dri/i965/brw_clip.c

@@ -150,7 +150,7 @@ static void upload_clip_prog(struct brw_context *brw)
    key.do_flat_shading = (ctx->Light.ShadeModel == GL_FLAT);
    key.pv_first = (ctx->Light.ProvokingVertex == GL_FIRST_VERTEX_CONVENTION);
    /* _NEW_TRANSFORM */
-   key.nr_userclip = brw_count_bits(ctx->Transform.ClipPlanesEnabled);
+   key.nr_userclip = _mesa_bitcount_64(ctx->Transform.ClipPlanesEnabled);
 
    if (intel->gen == 5)
        key.clip_mode = BRW_CLIPMODE_KERNEL_CLIP;

src/mesa/drivers/dri/i965/brw_curbe.c

@@ -66,7 +66,7 @@ static void calculate_curbe_offsets( struct brw_context *brw )
 
    /* _NEW_TRANSFORM */
    if (ctx->Transform.ClipPlanesEnabled) {
-      GLuint nr_planes = 6 + brw_count_bits(ctx->Transform.ClipPlanesEnabled);
+      GLuint nr_planes = 6 + _mesa_bitcount_64(ctx->Transform.ClipPlanesEnabled);
       nr_clip_regs = (nr_planes * 4 + 15) / 16;
    }
 

src/mesa/drivers/dri/i965/brw_gs.c

@@ -159,7 +159,7 @@ static void populate_key( struct brw_context *brw,
    }
 
    /* _NEW_TRANSFORM */
-   key->nr_userclip = brw_count_bits(ctx->Transform.ClipPlanesEnabled);
+   key->nr_userclip = _mesa_bitcount_64(ctx->Transform.ClipPlanesEnabled);
 
    key->need_gs_prog = (intel->gen >= 6)
       ? 0

src/mesa/drivers/dri/i965/brw_sf.c

@@ -154,7 +154,7 @@ static void upload_sf_prog(struct brw_context *brw)
    }
 
    /* _NEW_TRANSFORM */
-   key.nr_userclip = brw_count_bits(ctx->Transform.ClipPlanesEnabled);
+   key.nr_userclip = _mesa_bitcount_64(ctx->Transform.ClipPlanesEnabled);
 
    /* _NEW_POINT */
    key.do_point_sprite = ctx->Point.PointSprite;

src/mesa/drivers/dri/i965/brw_sf_emit.c

@@ -167,7 +167,7 @@ static void do_flatshade_triangle( struct brw_sf_compile *c )
    struct brw_compile *p = &c->func;
    struct intel_context *intel = &p->brw->intel;
    struct brw_reg ip = brw_ip_reg();
-   GLuint nr = brw_count_bits(c->key.attrs & VERT_RESULT_COLOR_BITS);
+   GLuint nr = _mesa_bitcount_64(c->key.attrs & VERT_RESULT_COLOR_BITS);
    GLuint jmpi = 1;
 
    if (!nr)
@@ -206,7 +206,7 @@ static void do_flatshade_line( struct brw_sf_compile *c )
    struct brw_compile *p = &c->func;
    struct intel_context *intel = &p->brw->intel;
    struct brw_reg ip = brw_ip_reg();
-   GLuint nr = brw_count_bits(c->key.attrs & VERT_RESULT_COLOR_BITS);
+   GLuint nr = _mesa_bitcount_64(c->key.attrs & VERT_RESULT_COLOR_BITS);
    GLuint jmpi = 1;
 
    if (!nr)

src/mesa/drivers/dri/i965/brw_util.h

@@ -34,15 +34,8 @@
 #define BRW_UTIL_H
 
 #include "main/mtypes.h"
+#include "main/imports.h"
 
-#ifdef __GNUC__
-#define brw_count_bits(v) __builtin_popcount(v)
-#else
-static inline GLuint brw_count_bits(uint64_t v)
-{
-	return _mesa_popcount(v>>32) + _mesa_popcount(v&0xffffffff);
-}
-#endif
 extern GLuint brw_parameter_list_state_flags(struct gl_program_parameter_list *paramList);
 extern GLuint brw_translate_blend_factor( GLenum factor );
 extern GLuint brw_translate_blend_equation( GLenum mode );

src/mesa/drivers/dri/i965/brw_vs.c

@@ -287,7 +287,7 @@ static void brw_upload_vs_prog(struct brw_context *brw)
     * the inputs it asks for, whether they are varying or not.
     */
    key.program_string_id = vp->id;
-   key.nr_userclip = brw_count_bits(ctx->Transform.ClipPlanesEnabled);
+   key.nr_userclip = _mesa_bitcount_64(ctx->Transform.ClipPlanesEnabled);
    key.uses_clip_distance = vp->program.UsesClipDistance;
    key.copy_edgeflag = (ctx->Polygon.FrontMode != GL_FILL ||
 			ctx->Polygon.BackMode != GL_FILL);

src/mesa/drivers/dri/i965/gen6_clip_state.c

@@ -46,7 +46,7 @@ brw_compute_userclip_flags(bool uses_clip_distance,
        * enable clipping planes 0 through n-1 in the hardware regardless of
        * which planes the user has selected.
        */
-      return (1 << brw_count_bits(clip_planes_enabled)) - 1;
+      return (1 << _mesa_bitcount_64(clip_planes_enabled)) - 1;
    }
 }
 

src/mesa/drivers/dri/i965/gen6_sf_state.c

@@ -117,7 +117,7 @@ upload_sf_state(struct brw_context *brw)
    /* CACHE_NEW_VS_PROG */
    GLbitfield64 vs_outputs_written = brw->vs.prog_data->outputs_written;
    /* BRW_NEW_FRAGMENT_PROGRAM */
-   uint32_t num_outputs = brw_count_bits(brw->fragment_program->Base.InputsRead);
+   uint32_t num_outputs = _mesa_bitcount_64(brw->fragment_program->Base.InputsRead);
    uint32_t dw1, dw2, dw3, dw4, dw16, dw17;
    int i;
    /* _NEW_BUFFER */
@@ -129,7 +129,7 @@ upload_sf_state(struct brw_context *brw)
    int nr_userclip;
 
    /* _NEW_TRANSFORM */
-   nr_userclip = brw_count_bits(ctx->Transform.ClipPlanesEnabled);
+   nr_userclip = _mesa_bitcount_64(ctx->Transform.ClipPlanesEnabled);
 
    brw_compute_vue_map(&vue_map, intel, nr_userclip, vs_outputs_written);
    urb_entry_read_length = (vue_map.num_slots + 1)/2 - urb_entry_read_offset;

src/mesa/drivers/dri/i965/gen6_wm_state.c

@@ -180,7 +180,7 @@ upload_wm_state(struct brw_context *brw)
 
    dw6 |= GEN6_WM_PERSPECTIVE_PIXEL_BARYCENTRIC;
 
-   dw6 |= brw_count_bits(brw->fragment_program->Base.InputsRead) <<
+   dw6 |= _mesa_bitcount_64(brw->fragment_program->Base.InputsRead) <<
       GEN6_WM_NUM_SF_OUTPUTS_SHIFT;
 
    BEGIN_BATCH(9);

src/mesa/drivers/dri/i965/gen7_sf_state.c

@@ -38,13 +38,13 @@ upload_sbe_state(struct brw_context *brw)
    /* CACHE_NEW_VS_PROG */
    GLbitfield64 vs_outputs_written = brw->vs.prog_data->outputs_written;
    /* BRW_NEW_FRAGMENT_PROGRAM */
-   uint32_t num_outputs = brw_count_bits(brw->fragment_program->Base.InputsRead);
+   uint32_t num_outputs = _mesa_bitcount_64(brw->fragment_program->Base.InputsRead);
    uint32_t dw1, dw10, dw11;
    int i;
    int attr = 0, input_index = 0;
    /* _NEW_TRANSFORM */
    int urb_entry_read_offset = 1;
-   int nr_userclip = brw_count_bits(ctx->Transform.ClipPlanesEnabled);
+   int nr_userclip = _mesa_bitcount_64(ctx->Transform.ClipPlanesEnabled);
    uint16_t attr_overrides[FRAG_ATTRIB_MAX];
 
    brw_compute_vue_map(&vue_map, intel, nr_userclip, vs_outputs_written);

src/mesa/main/imports.c

@@ -527,6 +527,19 @@ _mesa_bitcount(unsigned int n)
    }
    return bits;
 }
+
+/**
+ * Return number of bits set in given 64-bit uint.
+ */
+unsigned int
+_mesa_bitcount_64(uint64_t n)
+{
+   unsigned int bits;
+   for (bits = 0; n > 0; n = n >> 1) {
+      bits += (n & 1);
+   }
+   return bits;
+}
 #endif
 
 

src/mesa/main/imports.h

@@ -578,9 +578,12 @@ _mesa_init_sqrt_table(void);
 
 #if ((_GNUC__ == 3 && __GNUC_MINOR__ >= 4) || __GNUC__ >= 4)
 #define _mesa_bitcount(i) __builtin_popcount(i)
+#define _mesa_bitcount_64(i) __builtin_popcountll(i)
 #else
 extern unsigned int
 _mesa_bitcount(unsigned int n);
+extern unsigned int
+_mesa_bitcount_64(uint64_t n);
 #endif
 
 #else