gallium: change pipe_sampler_view::first_element/last_element -> offset/size

This is required by OpenGL. Our hardware supports this.

Example: Bind RGBA32F with offset = 4 bytes.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=97305

Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>
Acked-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

src/gallium/auxiliary/util/u_dump_state.c

@@ -750,8 +750,8 @@ util_dump_sampler_view(FILE *stream, const struct pipe_sampler_view *state)
    util_dump_member(stream, ptr, state, texture);
 
    if (state->target == PIPE_BUFFER) {
-      util_dump_member(stream, uint, state, u.buf.first_element);
+      util_dump_member(stream, uint, state, u.buf.offset);
-      util_dump_member(stream, uint, state, u.buf.last_element);
+      util_dump_member(stream, uint, state, u.buf.size);
    }
    else {
       util_dump_member(stream, uint, state, u.tex.first_layer);

src/gallium/docs/source/screen.rst

@@ -136,8 +136,8 @@ The integer capabilities:
   PIPE_BUFFER. In other words, the pointer returned by transfer_map is
   always aligned to this value.
 * ``PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT``: Describes the required
-  alignment for pipe_sampler_view::u.buf.first_element, in bytes.
+  alignment for pipe_sampler_view::u.buf.offset, in bytes.
-  If a driver does not support first/last_element, it should return 0.
+  If a driver does not support offset/size, it should return 0.
 * ``PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY``: Whether the driver only
   supports R, RG, RGB and RGBA formats for PIPE_BUFFER sampler views.
   When this is the case it should be assumed that the swizzle parameters

src/gallium/drivers/freedreno/a3xx/fd3_emit.c

@@ -206,8 +206,7 @@ emit_textures(struct fd_context *ctx, struct fd_ringbuffer *ring,
 					&dummy_view;
 			struct fd_resource *rsc = fd_resource(view->base.texture);
 			if (rsc && rsc->base.b.target == PIPE_BUFFER) {
-				OUT_RELOC(ring, rsc->bo, view->base.u.buf.first_element *
+				OUT_RELOC(ring, rsc->bo, view->base.u.buf.offset, 0, 0);
-						  util_format_get_blocksize(view->base.format), 0, 0);
 				j = 1;
 			} else {
 				unsigned start = fd_sampler_first_level(&view->base);

src/gallium/drivers/freedreno/a3xx/fd3_texture.c

@@ -241,8 +241,7 @@ fd3_sampler_view_create(struct pipe_context *pctx, struct pipe_resource *prsc,
 		lvl = 0;
 		so->texconst1 =
 			A3XX_TEX_CONST_1_FETCHSIZE(fd3_pipe2fetchsize(cso->format)) |
-			A3XX_TEX_CONST_1_WIDTH(cso->u.buf.last_element -
+			A3XX_TEX_CONST_1_WIDTH(cso->u.buf.size / util_format_get_blocksize(cso->format)) |
-								   cso->u.buf.first_element + 1) |
 			A3XX_TEX_CONST_1_HEIGHT(1);
 	} else {
 		unsigned miplevels;

src/gallium/drivers/freedreno/a4xx/fd4_texture.c

@@ -249,8 +249,8 @@ fd4_sampler_view_create(struct pipe_context *pctx, struct pipe_resource *prsc,
 	}
 
 	if (cso->target == PIPE_BUFFER) {
-		unsigned elements = cso->u.buf.last_element -
+		unsigned elements = cso->u.buf.size / util_format_get_blocksize(cso->format);
-			cso->u.buf.first_element + 1;
+
 		lvl = 0;
 		so->texconst1 =
 			A4XX_TEX_CONST_1_WIDTH(elements) |
@@ -258,8 +258,7 @@ fd4_sampler_view_create(struct pipe_context *pctx, struct pipe_resource *prsc,
 		so->texconst2 =
 			A4XX_TEX_CONST_2_FETCHSIZE(fd4_pipe2fetchsize(cso->format)) |
 			A4XX_TEX_CONST_2_PITCH(elements * rsc->cpp);
-		so->offset = cso->u.buf.first_element *
+		so->offset = cso->u.buf.offset;
-			util_format_get_blocksize(cso->format);
 	} else {
 		unsigned miplevels;
 

src/gallium/drivers/ilo/ilo_state.c

@@ -2015,9 +2015,8 @@ ilo_create_sampler_view(struct pipe_context *pipe,
 
       memset(&info, 0, sizeof(info));
       info.vma = ilo_resource_get_vma(res);
-      info.offset = templ->u.buf.first_element * info.struct_size;
+      info.offset = templ->u.buf.offset;
-      info.size = (templ->u.buf.last_element -
+      info.size = templ->u.buf.size;
-            templ->u.buf.first_element + 1) * info.struct_size;
       info.access = ILO_STATE_SURFACE_ACCESS_SAMPLER;
       info.format = ilo_format_translate_color(dev, templ->format);
       info.format_size = util_format_get_blocksize(templ->format);

src/gallium/drivers/llvmpipe/lp_setup.c

@@ -889,8 +889,8 @@ lp_setup_set_fragment_sampler_views(struct lp_setup_context *setup,
                }
                else {
                   /*
-                   * For buffers, we don't have first_element, instead adjust
+                   * For buffers, we don't have "offset", instead adjust
-                   * last_element (stored as width) plus the base pointer.
+                   * the size (stored as width) plus the base pointer.
                    */
                   unsigned view_blocksize = util_format_get_blocksize(view->format);
                   /* probably don't really need to fill that out */
@@ -899,12 +899,10 @@ lp_setup_set_fragment_sampler_views(struct lp_setup_context *setup,
                   jit_tex->img_stride[0] = 0;
 
                   /* everything specified in number of elements here. */
-                  jit_tex->width = view->u.buf.last_element - view->u.buf.first_element + 1;
+                  jit_tex->width = view->u.buf.size / view_blocksize;
-                  jit_tex->base = (uint8_t *)jit_tex->base + view->u.buf.first_element *
+                  jit_tex->base = (uint8_t *)jit_tex->base + view->u.buf.offset;
-                                  view_blocksize;
                   /* XXX Unsure if we need to sanitize parameters? */
-                  assert(view->u.buf.first_element <= view->u.buf.last_element);
+                  assert(view->u.buf.offset + view->u.buf.size <= res->width0);
-                  assert(view->u.buf.last_element * view_blocksize < res->width0);
                }
             }
          }

src/gallium/drivers/llvmpipe/lp_state_sampler.c

@@ -308,11 +308,9 @@ prepare_shader_sampling(
                img_stride[0] = 0;
 
                /* everything specified in number of elements here. */
-               width0 = view->u.buf.last_element - view->u.buf.first_element + 1;
+               width0 = view->u.buf.size / view_blocksize;
-               addr = (uint8_t *)addr + view->u.buf.first_element *
+               addr = (uint8_t *)addr + view->u.buf.offset;
-                               view_blocksize;
+               assert(view->u.buf.offset + view->u.buf.size <= res->width0);
-               assert(view->u.buf.first_element <= view->u.buf.last_element);
-               assert(view->u.buf.last_element * view_blocksize < res->width0);
             }
          }
          else {

src/gallium/drivers/nouveau/nv50/nv50_tex.c

@@ -131,11 +131,11 @@ nv50_create_texture_view(struct pipe_context *pipe,
 
    if (unlikely(!nouveau_bo_memtype(nv04_resource(texture)->bo))) {
       if (target == PIPE_BUFFER) {
-         addr += view->pipe.u.buf.first_element * desc->block.bits / 8;
+         addr += view->pipe.u.buf.offset;
          tic[2] |= G80_TIC_2_LAYOUT_PITCH | G80_TIC_2_TEXTURE_TYPE_ONE_D_BUFFER;
          tic[3] = 0;
          tic[4] = /* width */
-            view->pipe.u.buf.last_element - view->pipe.u.buf.first_element + 1;
+            view->pipe.u.buf.size / (desc->block.bits / 8);
          tic[5] = 0;
       } else {
          tic[2] |= G80_TIC_2_LAYOUT_PITCH | G80_TIC_2_TEXTURE_TYPE_TWO_D_NO_MIPMAP;
@@ -224,8 +224,7 @@ nv50_update_tic(struct nv50_context *nv50, struct nv50_tic_entry *tic,
    uint64_t address = res->address;
    if (res->base.target != PIPE_BUFFER)
       return;
-   address += tic->pipe.u.buf.first_element *
+   address += tic->pipe.u.buf.offset;
-      util_format_get_blocksize(tic->pipe.format);
    if (tic->tic[1] == (uint32_t)address &&
        (tic->tic[2] & 0xff) == address >> 32)
       return;

src/gallium/drivers/nouveau/nvc0/nvc0_tex.c

@@ -132,9 +132,9 @@ gm107_create_texture_view(struct pipe_context *pipe,
    if (unlikely(!nouveau_bo_memtype(nv04_resource(texture)->bo))) {
       if (texture->target == PIPE_BUFFER) {
          assert(!(tic[5] & GM107_TIC2_5_NORMALIZED_COORDS));
-         width = view->pipe.u.buf.last_element - view->pipe.u.buf.first_element;
+         width = view->pipe.u.buf.size / (desc->block.bits / 8) - 1;
          address +=
-            view->pipe.u.buf.first_element * desc->block.bits / 8;
+            view->pipe.u.buf.offset;
          tic[2]  = GM107_TIC2_2_HEADER_VERSION_ONE_D_BUFFER;
          tic[3] |= width >> 16;
          tic[4] |= GM107_TIC2_4_TEXTURE_TYPE_ONE_D_BUFFER;
@@ -259,8 +259,11 @@ gm107_create_texture_view_from_image(struct pipe_context *pipe,
    templ.swizzle_a = PIPE_SWIZZLE_W;
 
    if (target == PIPE_BUFFER) {
-      templ.u.buf.first_element = view->u.buf.first_element;
+      templ.u.buf.offset = view->u.buf.first_element *
-      templ.u.buf.last_element = view->u.buf.last_element;
+                           util_format_get_blocksize(view->format);
+      templ.u.buf.size = (view->u.buf.last_element -
+                          view->u.buf.first_element + 1) *
+                         util_format_get_blocksize(view->format);
    } else {
       templ.u.tex.first_layer = view->u.tex.first_layer;
       templ.u.tex.last_layer = view->u.tex.last_layer;
@@ -344,11 +347,11 @@ gf100_create_texture_view(struct pipe_context *pipe,
       if (texture->target == PIPE_BUFFER) {
          assert(!(tic[2] & G80_TIC_2_NORMALIZED_COORDS));
          address +=
-            view->pipe.u.buf.first_element * desc->block.bits / 8;
+            view->pipe.u.buf.offset;
          tic[2] |= G80_TIC_2_LAYOUT_PITCH | G80_TIC_2_TEXTURE_TYPE_ONE_D_BUFFER;
          tic[3] = 0;
          tic[4] = /* width */
-            view->pipe.u.buf.last_element - view->pipe.u.buf.first_element + 1;
+            view->pipe.u.buf.size / (desc->block.bits / 8);
          tic[5] = 0;
       } else {
          /* must be 2D texture without mip maps */
@@ -456,8 +459,7 @@ nvc0_update_tic(struct nvc0_context *nvc0, struct nv50_tic_entry *tic,
    uint64_t address = res->address;
    if (res->base.target != PIPE_BUFFER)
       return;
-   address += tic->pipe.u.buf.first_element *
+   address += tic->pipe.u.buf.offset;
-      util_format_get_blocksize(tic->pipe.format);
    if (tic->tic[1] == (uint32_t)address &&
        (tic->tic[2] & 0xff) == address >> 32)
       return;

src/gallium/drivers/r600/evergreen_state.c

@@ -613,8 +613,8 @@ texture_buffer_sampler_view(struct r600_context *rctx,
 	unsigned swizzle_res;
 	unsigned char swizzle[4];
 	const struct util_format_description *desc;
-	unsigned offset = view->base.u.buf.first_element * stride;
+	unsigned offset = view->base.u.buf.offset;
-	unsigned size = (view->base.u.buf.last_element - view->base.u.buf.first_element + 1) * stride;
+	unsigned size = view->base.u.buf.size;
 
 	swizzle[0] = view->base.swizzle_r;
 	swizzle[1] = view->base.swizzle_g;

src/gallium/drivers/r600/r600_state.c

@@ -631,8 +631,8 @@ texture_buffer_sampler_view(struct r600_pipe_sampler_view *view,
 	struct r600_texture *tmp = (struct r600_texture*)view->base.texture;
 	int stride = util_format_get_blocksize(view->base.format);
 	unsigned format, num_format, format_comp, endian;
-	uint64_t offset = view->base.u.buf.first_element * stride;
+	uint64_t offset = view->base.u.buf.offset;
-	unsigned size = (view->base.u.buf.last_element - view->base.u.buf.first_element + 1) * stride;
+	unsigned size = view->base.u.buf.size;
 
 	r600_vertex_data_type(view->base.format,
 			      &format, &num_format, &format_comp,

src/gallium/drivers/r600/r600_state_common.c

@@ -2831,8 +2831,7 @@ static void r600_invalidate_buffer(struct pipe_context *ctx, struct pipe_resourc
 	/* Texture buffer objects - update the virtual addresses in descriptors. */
 	LIST_FOR_EACH_ENTRY(view, &rctx->b.texture_buffers, list) {
 		if (view->base.texture == &rbuffer->b.b) {
-			unsigned stride = util_format_get_blocksize(view->base.format);
+			uint64_t offset = view->base.u.buf.offset;
-			uint64_t offset = (uint64_t)view->base.u.buf.first_element * stride;
 			uint64_t va = rbuffer->gpu_address + offset;
 
 			view->tex_resource_words[0] = va;

src/gallium/drivers/radeonsi/si_descriptors.c

@@ -641,8 +641,11 @@ static void si_set_shader_image(struct si_context *ctx,
 
 		si_make_buffer_descriptor(screen, res,
 					  view->format,
-					  view->u.buf.first_element,
+					  view->u.buf.first_element *
-					  view->u.buf.last_element,
+					  util_format_get_blocksize(view->format),
+					  (view->u.buf.last_element -
+					   view->u.buf.first_element + 1) *
+					  util_format_get_blocksize(view->format),
 					  descs->list + slot * 8);
 		images->compressed_colortex_mask &= ~(1 << slot);
 	} else {

src/gallium/drivers/radeonsi/si_state.c

@@ -2644,7 +2644,7 @@ static void si_set_min_samples(struct pipe_context *ctx, unsigned min_samples)
 void
 si_make_buffer_descriptor(struct si_screen *screen, struct r600_resource *buf,
 			  enum pipe_format format,
-			  unsigned first_element, unsigned last_element,
+			  unsigned offset, unsigned size,
 			  uint32_t *state)
 {
 	const struct util_format_description *desc;
@@ -2657,11 +2657,11 @@ si_make_buffer_descriptor(struct si_screen *screen, struct r600_resource *buf,
 	desc = util_format_description(format);
 	first_non_void = util_format_get_first_non_void_channel(format);
 	stride = desc->block.bits / 8;
-	va = buf->gpu_address + first_element * stride;
+	va = buf->gpu_address + offset;
 	num_format = si_translate_buffer_numformat(&screen->b.b, desc, first_non_void);
 	data_format = si_translate_buffer_dataformat(&screen->b.b, desc, first_non_void);
 
-	num_records = last_element + 1 - first_element;
+	num_records = size / stride;
 	num_records = MIN2(num_records, buf->b.b.width0 / stride);
 
 	if (screen->b.chip_class >= VI)
@@ -2960,8 +2960,8 @@ si_create_sampler_view_custom(struct pipe_context *ctx,
 		si_make_buffer_descriptor(sctx->screen,
 					  (struct r600_resource *)texture,
 					  state->format,
-					  state->u.buf.first_element,
+					  state->u.buf.offset,
-					  state->u.buf.last_element,
+					  state->u.buf.size,
 					  view->state);
 
 		LIST_ADDTAIL(&view->list, &sctx->b.texture_buffers);

src/gallium/drivers/radeonsi/si_state.h

@@ -313,7 +313,7 @@ void si_init_screen_state_functions(struct si_screen *sscreen);
 void
 si_make_buffer_descriptor(struct si_screen *screen, struct r600_resource *buf,
 			  enum pipe_format format,
-			  unsigned first_element, unsigned last_element,
+			  unsigned offset, unsigned size,
 			  uint32_t *state);
 void
 si_make_texture_descriptor(struct si_screen *screen,

src/gallium/drivers/softpipe/sp_state_sampler.c

@@ -240,11 +240,9 @@ prepare_shader_sampling(
                img_stride[0] = 0;
 
                /* everything specified in number of elements here. */
-               width0 = view->u.buf.last_element - view->u.buf.first_element + 1;
+               width0 = view->u.buf.size / view_blocksize;
-               addr = (uint8_t *)addr + view->u.buf.first_element *
+               addr = (uint8_t *)addr + view->u.buf.offset;
-                               view_blocksize;
+               assert(view->u.buf.offset + view->u.buf.size <= res->width0);
-               assert(view->u.buf.first_element <= view->u.buf.last_element);
-               assert(view->u.buf.last_element * view_blocksize < res->width0);
             }
          }
          else {

src/gallium/drivers/softpipe/sp_tex_sample.c

@@ -3192,7 +3192,7 @@ sp_get_dims(const struct sp_sampler_view *sp_sview,
    const struct pipe_resource *texture = view->texture;
 
    if (view->target == PIPE_BUFFER) {
-      dims[0] = (view->u.buf.last_element - view->u.buf.first_element) + 1;
+      dims[0] = view->u.buf.size / util_format_get_blocksize(view->format);
       /* the other values are undefined, but let's avoid potential valgrind
        * warnings.
        */
@@ -3264,17 +3264,22 @@ sp_get_texels(const struct sp_sampler_view *sp_sview,
    const int width = u_minify(texture->width0, level);
    const int height = u_minify(texture->height0, level);
    const int depth = u_minify(texture->depth0, level);
+   unsigned elem_size, first_element, last_element;
 
    addr.value = 0;
    addr.bits.level = level;
 
    switch (sp_sview->base.target) {
    case PIPE_BUFFER:
+      elem_size = util_format_get_blocksize(sp_sview->base.format);
+      first_element = sp_sview->base.u.buf.offset / elem_size;
+      last_element = (sp_sview->base.u.buf.offset +
+                      sp_sview->base.u.buf.size) / elem_size - 1;
       for (j = 0; j < TGSI_QUAD_SIZE; j++) {
          const int x = CLAMP(v_i[j] + offset[0] +
-                             sp_sview->base.u.buf.first_element,
+                             first_element,
-                             sp_sview->base.u.buf.first_element,
+                             first_element,
-                             sp_sview->base.u.buf.last_element);
+                             last_element);
          tx = get_texel_2d_no_border(sp_sview, addr, x, 0);
          for (c = 0; c < 4; c++) {
             rgba[c][j] = tx[c];

src/gallium/drivers/svga/svga_state_sampler.c

@@ -30,6 +30,7 @@
 
 #include "pipe/p_defines.h"
 #include "util/u_bitmask.h"
+#include "util/u_format.h"
 #include "util/u_inlines.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
@@ -125,9 +126,10 @@ svga_validate_pipe_sampler_view(struct svga_context *svga,
       format = svga_sampler_format(format);
 
       if (texture->target == PIPE_BUFFER) {
-         viewDesc.buffer.firstElement = sv->base.u.buf.first_element;
+         unsigned elem_size = util_format_get_blocksize(sv->base.format);
-         viewDesc.buffer.numElements = (sv->base.u.buf.last_element -
+
-                                        sv->base.u.buf.first_element + 1);
+         viewDesc.buffer.firstElement = sv->base.u.buf.offset / elem_size;
+         viewDesc.buffer.numElements = sv->base.u.buf.size / elem_size;
       }
       else {
          viewDesc.tex.mostDetailedMip = sv->base.u.tex.first_level;

src/gallium/drivers/trace/tr_dump_state.c

@@ -543,8 +543,8 @@ void trace_dump_sampler_view_template(const struct pipe_sampler_view *state,
    if (target == PIPE_BUFFER) {
       trace_dump_member_begin("buf");
       trace_dump_struct_begin(""); /* anonymous */
-      trace_dump_member(uint, &state->u.buf, first_element);
+      trace_dump_member(uint, &state->u.buf, offset);
-      trace_dump_member(uint, &state->u.buf, last_element);
+      trace_dump_member(uint, &state->u.buf, size);
       trace_dump_struct_end(); /* anonymous */
       trace_dump_member_end(); /* buf */
    } else {

src/gallium/drivers/virgl/virgl_encode.c

@@ -22,6 +22,7 @@
  */
 #include <stdint.h>
 
+#include "util/u_format.h"
 #include "util/u_memory.h"
 #include "util/u_math.h"
 #include "pipe/p_state.h"
@@ -562,14 +563,16 @@ int virgl_encode_sampler_view(struct virgl_context *ctx,
                              struct virgl_resource *res,
                              const struct pipe_sampler_view *state)
 {
+   unsigned elem_size = util_format_get_blocksize(state->format);
+
    uint32_t tmp;
    virgl_encoder_write_cmd_dword(ctx, VIRGL_CMD0(VIRGL_CCMD_CREATE_OBJECT, VIRGL_OBJECT_SAMPLER_VIEW, VIRGL_OBJ_SAMPLER_VIEW_SIZE));
    virgl_encoder_write_dword(ctx->cbuf, handle);
    virgl_encoder_write_res(ctx, res);
    virgl_encoder_write_dword(ctx->cbuf, state->format);
    if (res->u.b.target == PIPE_BUFFER) {
-      virgl_encoder_write_dword(ctx->cbuf, state->u.buf.first_element);
+      virgl_encoder_write_dword(ctx->cbuf, state->u.buf.offset / elem_size);
-      virgl_encoder_write_dword(ctx->cbuf, state->u.buf.last_element);
+      virgl_encoder_write_dword(ctx->cbuf, (state->u.buf.offset + state->u.buf.size) / elem_size - 1);
    } else {
       virgl_encoder_write_dword(ctx->cbuf, state->u.tex.first_layer | state->u.tex.last_layer << 16);
       virgl_encoder_write_dword(ctx->cbuf, state->u.tex.first_level | state->u.tex.last_level << 8);

src/gallium/include/pipe/p_state.h

@@ -428,8 +428,8 @@ struct pipe_sampler_view
          unsigned last_level:8;    /**< last mipmap level to use */
       } tex;
       struct {
-         unsigned first_element;
+         unsigned offset;   /**< offset in bytes */
-         unsigned last_element;
+         unsigned size;     /**< size of the readable sub-range in bytes */
       } buf;
    } u;
    unsigned swizzle_r:3;         /**< PIPE_SWIZZLE_x for red component */

src/mesa/state_tracker/st_atom_texture.c

@@ -279,21 +279,16 @@ st_create_texture_sampler_view_from_stobj(struct st_context *st,
 
    if (stObj->pt->target == PIPE_BUFFER) {
       unsigned base, size;
-      unsigned f, n;
-      const struct util_format_description *desc
-         = util_format_description(templ.format);
 
       base = stObj->base.BufferOffset;
       if (base >= stObj->pt->width0)
          return NULL;
       size = MIN2(stObj->pt->width0 - base, (unsigned)stObj->base.BufferSize);
-
+      if (!size)
-      f = (base / (desc->block.bits / 8)) * desc->block.width;
-      n = (size / (desc->block.bits / 8)) * desc->block.width;
-      if (!n)
          return NULL;
-      templ.u.buf.first_element = f;
+
-      templ.u.buf.last_element  = f + (n - 1);
+      templ.u.buf.offset = base;
+      templ.u.buf.size = size;
    } else {
       templ.u.tex.first_level = stObj->base.MinLevel + stObj->base.BaseLevel;
       templ.u.tex.last_level = last_level(stObj);

src/mesa/state_tracker/st_cb_texture.c

@@ -1163,8 +1163,9 @@ try_pbo_upload_common(struct gl_context *ctx,
       memset(&templ, 0, sizeof(templ));
       templ.target = PIPE_BUFFER;
       templ.format = src_format;
-      templ.u.buf.first_element = addr->first_element;
+      templ.u.buf.offset = addr->first_element * addr->bytes_per_pixel;
-      templ.u.buf.last_element = addr->last_element;
+      templ.u.buf.size = (addr->last_element - addr->first_element + 1) *
+                         addr->bytes_per_pixel;
       templ.swizzle_r = PIPE_SWIZZLE_X;
       templ.swizzle_g = PIPE_SWIZZLE_Y;
       templ.swizzle_b = PIPE_SWIZZLE_Z;