draw: don't assume vertex position is in data[0]

diff --git a/src/gallium/auxiliary/draw/draw_pipe_aaline.c b/src/gallium/auxiliary/draw/draw_pipe_aaline.c
index 634bf067f147338fe538142b28c40975500380f6..ecdebca5f1270bc9d1a7967b5f19e289dcc810d8 100644 (file)
--- a/src/gallium/auxiliary/draw/draw_pipe_aaline.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_aaline.c
@@ -78,6 +78,8 @@ struct aaline_stage
 
    /** For AA lines, this is the vertex attrib slot for the new texcoords */
    uint tex_slot;
+   /** position, not necessarily output zero */
+   uint pos_slot;
 
    void *sampler_cso;
    struct pipe_texture *texture;
@@ -520,9 +522,10 @@ aaline_line(struct draw_stage *stage, struct prim_header *header)
    struct prim_header tri;
    struct vertex_header *v[8];
    uint texPos = aaline->tex_slot;
+   uint posPos = aaline->pos_slot;
    float *pos, *tex;
-   float dx = header->v[1]->data[0][0] - header->v[0]->data[0][0];
-   float dy = header->v[1]->data[0][1] - header->v[0]->data[0][1];
+   float dx = header->v[1]->data[posPos][0] - header->v[0]->data[posPos][0];
+   float dy = header->v[1]->data[posPos][1] - header->v[0]->data[posPos][1];
    double a = atan2(dy, dx);
    float c_a = (float) cos(a), s_a = (float) sin(a);
    uint i;
@@ -549,35 +552,35 @@ aaline_line(struct draw_stage *stage, struct prim_header *header)
     */
 
    /* new verts */
-   pos = v[0]->data[0];
+   pos = v[0]->data[posPos];
    pos[0] += (-dx * c_a -  dy * s_a);
    pos[1] += (-dx * s_a +  dy * c_a);
 
-   pos = v[1]->data[0];
+   pos = v[1]->data[posPos];
    pos[0] += (-dx * c_a - -dy * s_a);
    pos[1] += (-dx * s_a + -dy * c_a);
 
-   pos = v[2]->data[0];
+   pos = v[2]->data[posPos];
    pos[0] += ( dx * c_a -  dy * s_a);
    pos[1] += ( dx * s_a +  dy * c_a);
 
-   pos = v[3]->data[0];
+   pos = v[3]->data[posPos];
    pos[0] += ( dx * c_a - -dy * s_a);
    pos[1] += ( dx * s_a + -dy * c_a);
 
-   pos = v[4]->data[0];
+   pos = v[4]->data[posPos];
    pos[0] += (-dx * c_a -  dy * s_a);
    pos[1] += (-dx * s_a +  dy * c_a);
 
-   pos = v[5]->data[0];
+   pos = v[5]->data[posPos];
    pos[0] += (-dx * c_a - -dy * s_a);
    pos[1] += (-dx * s_a + -dy * c_a);
 
-   pos = v[6]->data[0];
+   pos = v[6]->data[posPos];
    pos[0] += ( dx * c_a -  dy * s_a);
    pos[1] += ( dx * s_a +  dy * c_a);
 
-   pos = v[7]->data[0];
+   pos = v[7]->data[posPos];
    pos[0] += ( dx * c_a - -dy * s_a);
    pos[1] += ( dx * s_a + -dy * c_a);
 
@@ -653,7 +656,7 @@ aaline_first_line(struct draw_stage *stage, struct prim_header *header)
 
    /* update vertex attrib info */
    aaline->tex_slot = draw->vs.num_vs_outputs;
-   assert(aaline->tex_slot > 0); /* output[0] is vertex pos */
+   aaline->pos_slot = draw->vs.position_output;
 
    /* advertise the extra post-transformed vertex attribute */
    draw->extra_vp_outputs.semantic_name = TGSI_SEMANTIC_GENERIC;

diff --git a/src/gallium/auxiliary/draw/draw_pipe_aapoint.c b/src/gallium/auxiliary/draw/draw_pipe_aapoint.c
index 96dcdb43d54d37e785c4814ce9fb271f410447d6..87fd30364934ffd9cf6b081e37e6057a7737b631 100644 (file)
--- a/src/gallium/auxiliary/draw/draw_pipe_aapoint.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_aapoint.c
@@ -85,6 +85,7 @@ struct aapoint_stage
 
    /** this is the vertex attrib slot for the new texcoords */
    uint tex_slot;
+   uint pos_slot;
 
    /*
     * Currently bound state
@@ -570,6 +571,7 @@ aapoint_point(struct draw_stage *stage, struct prim_header *header)
    struct prim_header tri;
    struct vertex_header *v[4];
    uint texPos = aapoint->tex_slot;
+   uint pos_slot = aapoint->pos_slot;
    float radius, *pos, *tex;
    uint i;
    float k;
@@ -619,19 +621,19 @@ aapoint_point(struct draw_stage *stage, struct prim_header *header)
    }
 
    /* new verts */
-   pos = v[0]->data[0];
+   pos = v[0]->data[pos_slot];
    pos[0] -= radius;
    pos[1] -= radius;
 
-   pos = v[1]->data[0];
+   pos = v[1]->data[pos_slot];
    pos[0] += radius;
    pos[1] -= radius;
 
-   pos = v[2]->data[0];
+   pos = v[2]->data[pos_slot];
    pos[0] += radius;
    pos[1] += radius;
 
-   pos = v[3]->data[0];
+   pos = v[3]->data[pos_slot];
    pos[0] -= radius;
    pos[1] += radius;
 
@@ -683,6 +685,8 @@ aapoint_first_point(struct draw_stage *stage, struct prim_header *header)
    aapoint->tex_slot = draw->vs.num_vs_outputs;
    assert(aapoint->tex_slot > 0); /* output[0] is vertex pos */
 
+   aapoint->pos_slot = draw->vs.position_output;
+
    draw->extra_vp_outputs.semantic_name = TGSI_SEMANTIC_GENERIC;
    draw->extra_vp_outputs.semantic_index = aapoint->fs->generic_attrib;
    draw->extra_vp_outputs.slot = aapoint->tex_slot;

diff --git a/src/gallium/auxiliary/draw/draw_pipe_clip.c b/src/gallium/auxiliary/draw/draw_pipe_clip.c
index 77ccddac4abfb85733fd2cb0adcffd8d9b883040..fa10f8efca2c61c3b4fd0b012fe9c3b5d7c5e234 100644 (file)
--- a/src/gallium/auxiliary/draw/draw_pipe_clip.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_clip.c
@@ -113,6 +113,7 @@ static void interp( const struct clipper *clip,
                    const struct vertex_header *in )
 {
    const unsigned nr_attrs = clip->stage.draw->vs.num_vs_outputs;
+   const unsigned pos_attr = clip->stage.draw->vs.position_output;
    unsigned j;
 
    /* Vertex header.
@@ -138,18 +139,17 @@ static void interp( const struct clipper *clip,
       const float *trans = clip->stage.draw->viewport.translate;
       const float oow = 1.0f / pos[3];
 
-      dst->data[0][0] = pos[0] * oow * scale[0] + trans[0];
-      dst->data[0][1] = pos[1] * oow * scale[1] + trans[1];
-      dst->data[0][2] = pos[2] * oow * scale[2] + trans[2];
-      dst->data[0][3] = oow;
+      dst->data[pos_attr][0] = pos[0] * oow * scale[0] + trans[0];
+      dst->data[pos_attr][1] = pos[1] * oow * scale[1] + trans[1];
+      dst->data[pos_attr][2] = pos[2] * oow * scale[2] + trans[2];
+      dst->data[pos_attr][3] = oow;
    }
 
    /* Other attributes
-    * Note: start at 1 to skip winpos (data[0]) since we just computed
-    * it above.
     */
-   for (j = 1; j < nr_attrs; j++) {
-      interp_attr(dst->data[j], t, in->data[j], out->data[j]);
+   for (j = 0; j < nr_attrs; j++) {
+      if (j != pos_attr)
+         interp_attr(dst->data[j], t, in->data[j], out->data[j]);
    }
 }
 

diff --git a/src/gallium/auxiliary/draw/draw_pipe_cull.c b/src/gallium/auxiliary/draw/draw_pipe_cull.c
index 87aaf1f85bd0c0add4b40e00308324c7cb9cbea2..d0d22a38e07be8b7f30283ecd4dfc437d16f20b0 100644 (file)
--- a/src/gallium/auxiliary/draw/draw_pipe_cull.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_cull.c
@@ -55,10 +55,12 @@ static INLINE struct cull_stage *cull_stage( struct draw_stage *stage )
 static void cull_tri( struct draw_stage *stage,
                      struct prim_header *header )
 {
+   const unsigned pos = stage->draw->vs.position_output;
+
    /* Window coords: */
-   const float *v0 = header->v[0]->data[0];
-   const float *v1 = header->v[1]->data[0];
-   const float *v2 = header->v[2]->data[0];
+   const float *v0 = header->v[0]->data[pos];
+   const float *v1 = header->v[1]->data[pos];
+   const float *v2 = header->v[2]->data[pos];
 
    /* edge vectors e = v0 - v2, f = v1 - v2 */
    const float ex = v0[0] - v2[0];

diff --git a/src/gallium/auxiliary/draw/draw_pipe_offset.c b/src/gallium/auxiliary/draw/draw_pipe_offset.c
index ea6de8c571eb1e203c9085352eaa8376705afa7c..8f1650e55ca25fbf6c189064f0dd71dfeae85354 100644 (file)
--- a/src/gallium/auxiliary/draw/draw_pipe_offset.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_offset.c
@@ -62,14 +62,15 @@ static INLINE struct offset_stage *offset_stage( struct draw_stage *stage )
 static void do_offset_tri( struct draw_stage *stage,
                           struct prim_header *header )
 {
+   const unsigned pos = stage->draw->vs.position_output;
    struct offset_stage *offset = offset_stage(stage);   
    float inv_det = 1.0f / header->det;
 
    /* Window coords:
     */
-   float *v0 = header->v[0]->data[0];
-   float *v1 = header->v[1]->data[0];
-   float *v2 = header->v[2]->data[0];
+   float *v0 = header->v[0]->data[pos];
+   float *v1 = header->v[1]->data[pos];
+   float *v2 = header->v[2]->data[pos];
 
    /* edge vectors e = v0 - v2, f = v1 - v2 */
    float ex = v0[0] - v2[0];

diff --git a/src/gallium/auxiliary/draw/draw_pipe_stipple.c b/src/gallium/auxiliary/draw/draw_pipe_stipple.c
index 9522b79582eaef6de91f8f3d9e3f9a168be7f4e1..bf0db18a684fb89ee40989371c8037fdb7c603a6 100644 (file)
--- a/src/gallium/auxiliary/draw/draw_pipe_stipple.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_stipple.c
@@ -119,8 +119,9 @@ stipple_line(struct draw_stage *stage, struct prim_header *header)
    struct stipple_stage *stipple = stipple_stage(stage);
    struct vertex_header *v0 = header->v[0];
    struct vertex_header *v1 = header->v[1];
-   const float *pos0 = v0->data[0];
-   const float *pos1 = v1->data[0];
+   const unsigned pos = stage->draw->vs.position_output;
+   const float *pos0 = v0->data[pos];
+   const float *pos1 = v1->data[pos];
    float start = 0;
    int state = 0;
 

diff --git a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
index 9e5597c32c44438e995881cf265cdc2e76539f70..10a1f7df7990377bd05bee26033c37dccc843f06 100644 (file)
--- a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
@@ -136,6 +136,8 @@ emit_vertex( struct vbuf_stage *vbuf,
        * set_buffer is efficient.  Consider a special one-shot mode for
        * translate.
        */
+      /* Note: we really do want data[0] here, not data[pos]: 
+       */
       vbuf->translate->set_buffer(vbuf->translate, 0, vertex->data[0], 0);
       vbuf->translate->run(vbuf->translate, 0, 1, vbuf->vertex_ptr);
 

diff --git a/src/gallium/auxiliary/draw/draw_pipe_wide_line.c b/src/gallium/auxiliary/draw/draw_pipe_wide_line.c
index 878c9c716985868481878d49efebf8f1ae53fdd5..29649f57879debbda8cc7669f92866af2eff72fa 100644 (file)
--- a/src/gallium/auxiliary/draw/draw_pipe_wide_line.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_wide_line.c
@@ -58,6 +58,7 @@ static void wideline_line( struct draw_stage *stage,
                            struct prim_header *header )
 {
    /*const struct wideline_stage *wide = wideline_stage(stage);*/
+   const unsigned pos = stage->draw->vs.position_output;
    const float half_width = 0.5f * stage->draw->rasterizer->line_width;
 
    struct prim_header tri;
@@ -67,10 +68,10 @@ static void wideline_line( struct draw_stage *stage,
    struct vertex_header *v2 = dup_vert(stage, header->v[1], 2);
    struct vertex_header *v3 = dup_vert(stage, header->v[1], 3);
 
-   float *pos0 = v0->data[0];
-   float *pos1 = v1->data[0];
-   float *pos2 = v2->data[0];
-   float *pos3 = v3->data[0];
+   float *pos0 = v0->data[pos];
+   float *pos1 = v1->data[pos];
+   float *pos2 = v2->data[pos];
+   float *pos3 = v3->data[pos];
 
    const float dx = FABSF(pos0[0] - pos2[0]);
    const float dy = FABSF(pos0[1] - pos2[1]);

diff --git a/src/gallium/auxiliary/draw/draw_pipe_wide_point.c b/src/gallium/auxiliary/draw/draw_pipe_wide_point.c
index df92e3f2d06fff6c96db162406cc73a2d1b3832c..d40a07f4aed9422788e3c60be39788417e2dda01 100644 (file)
--- a/src/gallium/auxiliary/draw/draw_pipe_wide_point.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_wide_point.c
@@ -96,6 +96,7 @@ static void widepoint_point( struct draw_stage *stage,
                              struct prim_header *header )
 {
    const struct widepoint_stage *wide = widepoint_stage(stage);
+   const unsigned pos = stage->draw->vs.position_output;
    const boolean sprite = (boolean) stage->draw->rasterizer->point_sprite;
    float half_size;
    float left_adj, right_adj, bot_adj, top_adj;
@@ -108,10 +109,10 @@ static void widepoint_point( struct draw_stage *stage,
    struct vertex_header *v2 = dup_vert(stage, header->v[0], 2);
    struct vertex_header *v3 = dup_vert(stage, header->v[0], 3);
 
-   float *pos0 = v0->data[0];
-   float *pos1 = v1->data[0];
-   float *pos2 = v2->data[0];
-   float *pos3 = v3->data[0];
+   float *pos0 = v0->data[pos];
+   float *pos1 = v1->data[pos];
+   float *pos2 = v2->data[pos];
+   float *pos3 = v3->data[pos];
 
    /* point size is either per-vertex or fixed size */
    if (wide->psize_slot >= 0) {

diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h
index 1865601cc048559c01df4532c4507111f34b8906..7bd1e670b4b9120f93bde2be07b5b18cf4bfe431 100644 (file)
--- a/src/gallium/auxiliary/draw/draw_private.h
+++ b/src/gallium/auxiliary/draw/draw_private.h
@@ -177,7 +177,7 @@ struct draw_context
    struct {
       struct draw_vertex_shader *vertex_shader;
       uint num_vs_outputs;  /**< convenience, from vertex_shader */
-
+      uint position_output;
 
       /** TGSI program interpreter runtime state */
       struct tgsi_exec_machine machine;

diff --git a/src/gallium/auxiliary/draw/draw_pt_post_vs.c b/src/gallium/auxiliary/draw/draw_pt_post_vs.c
index c4a67c8289dc500f777884b0ba6ffcb7b728e559..af6306b1c67ccc0833369187cee70cda04595d61 100644 (file)
--- a/src/gallium/auxiliary/draw/draw_pt_post_vs.c
+++ b/src/gallium/auxiliary/draw/draw_pt_post_vs.c
@@ -100,16 +100,19 @@ static boolean post_vs_cliptest_viewport_gl( struct pt_post_vs *pvs,
    struct vertex_header *out = vertices;
    const float *scale = pvs->draw->viewport.scale;
    const float *trans = pvs->draw->viewport.translate;
+   const unsigned pos = pvs->draw->vs.position_output;
    unsigned clipped = 0;
    unsigned j;
 
    if (0) debug_printf("%s\n");
 
    for (j = 0; j < count; j++) {
-      out->clip[0] = out->data[0][0];
-      out->clip[1] = out->data[0][1];
-      out->clip[2] = out->data[0][2];
-      out->clip[3] = out->data[0][3];
+      float *position = out->data[pos];
+
+      out->clip[0] = position[0];
+      out->clip[1] = position[1];
+      out->clip[2] = position[2];
+      out->clip[3] = position[3];
 
       out->vertex_id = 0xffff;
       out->clipmask = compute_clipmask_gl(out->clip, 
@@ -120,19 +123,19 @@ static boolean post_vs_cliptest_viewport_gl( struct pt_post_vs *pvs,
       if (out->clipmask == 0)
       {
         /* divide by w */
-        float w = 1.0f / out->data[0][3];
+        float w = 1.0f / position[3];
 
         /* Viewport mapping */
-        out->data[0][0] = out->data[0][0] * w * scale[0] + trans[0];
-        out->data[0][1] = out->data[0][1] * w * scale[1] + trans[1];
-        out->data[0][2] = out->data[0][2] * w * scale[2] + trans[2];
-        out->data[0][3] = w;
+        position[0] = position[0] * w * scale[0] + trans[0];
+        position[1] = position[1] * w * scale[1] + trans[1];
+        position[2] = position[2] * w * scale[2] + trans[2];
+        position[3] = w;
 #if 0
          debug_printf("post viewport: %f %f %f %f\n",
-                      out->data[0][0],
-                      out->data[0][1],
-                      out->data[0][2],
-                      out->data[0][3]);
+                      position[0],
+                      position[1],
+                      position[2],
+                      position[3]);
 #endif
       }
 
@@ -154,15 +157,18 @@ static boolean post_vs_viewport( struct pt_post_vs *pvs,
    struct vertex_header *out = vertices;
    const float *scale = pvs->draw->viewport.scale;
    const float *trans = pvs->draw->viewport.translate;
+   const unsigned pos = pvs->draw->vs.position_output;
    unsigned j;
 
    if (0) debug_printf("%s\n", __FUNCTION__);
    for (j = 0; j < count; j++) {
+      float *position = out->data[pos];
+
       /* Viewport mapping only, no cliptest/rhw divide
        */
-      out->data[0][0] = out->data[0][0] * scale[0] + trans[0];
-      out->data[0][1] = out->data[0][1] * scale[1] + trans[1];
-      out->data[0][2] = out->data[0][2] * scale[2] + trans[2];
+      position[0] = position[0] * scale[0] + trans[0];
+      position[1] = position[1] * scale[1] + trans[1];
+      position[2] = position[2] * scale[2] + trans[2];
 
       out = (struct vertex_header *)((char *)out + stride);
    }

diff --git a/src/gallium/auxiliary/draw/draw_vs.c b/src/gallium/auxiliary/draw/draw_vs.c
index 979f9864fdb0b496b491fd64b3315d53b8386054..978954e91c979b99f78702bc3c6d8f10b54f21b1 100644 (file)
--- a/src/gallium/auxiliary/draw/draw_vs.c
+++ b/src/gallium/auxiliary/draw/draw_vs.c
@@ -85,6 +85,16 @@ draw_create_vertex_shader(struct draw_context *draw,
       }
    }
 
+   if (vs)
+   {
+      uint i;
+      for (i = 0; i < vs->info.num_outputs; i++) {
+         if (vs->info.output_semantic_name[i] == TGSI_SEMANTIC_POSITION &&
+             vs->info.output_semantic_index[i] == 0)
+            vs->position_output = i;
+      }
+   }
+
    assert(vs);
    return vs;
 }
@@ -100,6 +110,7 @@ draw_bind_vertex_shader(struct draw_context *draw,
    {
       draw->vs.vertex_shader = dvs;
       draw->vs.num_vs_outputs = dvs->info.num_outputs;
+      draw->vs.position_output = dvs->position_output;
       dvs->prepare( dvs, draw );
    }
    else {

diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h
index 08c6de8ba86218e6f03c5146896e222f4df4a277..45992d19867c311038ce36377ebd08aff61d2814 100644 (file)
--- a/src/gallium/auxiliary/draw/draw_vs.h
+++ b/src/gallium/auxiliary/draw/draw_vs.h
@@ -106,6 +106,7 @@ struct draw_vertex_shader {
    struct pipe_shader_state   state;
 
    struct tgsi_shader_info info;
+   unsigned position_output;
 
    /* Extracted from shader:
     */

diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
index c47647ea72dc351ec7c8c3089fe9fa7a7324cdc0..5d4a8b38c876f133b37a75dab77683291b188f46 100644 (file)
--- a/src/gallium/auxiliary/draw/draw_vs_aos.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -1686,7 +1686,7 @@ static boolean emit_viewport( struct aos_compilation *cp )
 {
    struct x86_reg pos = aos_get_shader_reg_xmm(cp, 
                                                TGSI_FILE_OUTPUT, 
-                                               0);
+                                               cp->vaos->draw->vs.position_output );
 
    struct x86_reg scale = x86_make_disp(cp->machine_EDX, 
                                         Offset(struct aos_machine, scale));
@@ -1700,7 +1700,7 @@ static boolean emit_viewport( struct aos_compilation *cp )
    aos_adopt_xmm_reg( cp,
                       pos,
                       TGSI_FILE_OUTPUT,
-                      0,
+                      cp->vaos->draw->vs.position_output,
                       TRUE );
    return TRUE;
 }
@@ -1715,7 +1715,7 @@ static boolean emit_rhw_viewport( struct aos_compilation *cp )
    struct x86_reg tmp = aos_get_xmm_reg(cp);
    struct x86_reg pos = aos_get_shader_reg_xmm(cp, 
                                                TGSI_FILE_OUTPUT, 
-                                               0);
+                                               cp->vaos->draw->vs.position_output);
 
    struct x86_reg scale = x86_make_disp(cp->machine_EDX, 
                                         Offset(struct aos_machine, scale));
@@ -1740,7 +1740,7 @@ static boolean emit_rhw_viewport( struct aos_compilation *cp )
    aos_adopt_xmm_reg( cp,
                       pos,
                       TGSI_FILE_OUTPUT,
-                      0,
+                      cp->vaos->draw->vs.position_output,
                       TRUE );
    return TRUE;
 }

diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c
index abe8c5ec2dd27b7f7908069f2420bda4cb09fb3f..ad0b829afa1d0ca8b1229c33778d9d752c3a500b 100644 (file)
--- a/src/gallium/auxiliary/draw/draw_vs_varient.c
+++ b/src/gallium/auxiliary/draw/draw_vs_varient.c
@@ -89,6 +89,8 @@ static void do_rhw_viewport( struct draw_vs_varient_generic *vsvg,
    unsigned stride = vsvg->temp_vertex_stride;
    unsigned j;
 
+   ptr += vsvg->base.vs->position_output * 4 * sizeof(float);
+
    for (j = 0; j < count; j++, ptr += stride) {
       float *data = (float *)ptr;
       float w = 1.0f / data[3];
@@ -110,6 +112,8 @@ static void do_viewport( struct draw_vs_varient_generic *vsvg,
    unsigned stride = vsvg->temp_vertex_stride;
    unsigned j;
 
+   ptr += vsvg->base.vs->position_output * 4 * sizeof(float);
+
    for (j = 0; j < count; j++, ptr += stride) {
       float *data = (float *)ptr;