softpipe: add image support to softpipe (v3)

This adds support for ARB_shader_image_load_store to softpipe.

v2: add RESQ support (Ilia)
v3: constify, cleanup internals, add some comments (Brian).

Reviewed-by: Brian Paul <[email protected]>
Signed-off-by: Dave Airlie <[email protected]>

1 files changed, 762 insertions, 0 deletions

diff --git a/src/gallium/drivers/softpipe/sp_image.c b/src/gallium/drivers/softpipe/sp_image.c
new file mode 100644
index 00000000000..3488fa83185
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_image.c
@@ -0,0 +1,762 @@
+/*
+ * Copyright 2016 Red Hat.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#include "sp_context.h"
+#include "sp_image.h"
+#include "sp_texture.h"
+
+#include "util/u_format.h"
+
+/*
+ * Get the offset into the base image
+ * first element for a buffer or layer/level for texture.
+ */
+static uint32_t
+get_image_offset(const struct softpipe_resource *spr,
+                 const struct pipe_image_view *iview,
+                 enum pipe_format format, unsigned r_coord)
+{
+   int base_layer = 0;
+
+   if (spr->base.target == PIPE_BUFFER)
+      return iview->u.buf.first_element * util_format_get_blocksize(format);
+
+   if (spr->base.target == PIPE_TEXTURE_1D_ARRAY ||
+       spr->base.target == PIPE_TEXTURE_2D_ARRAY ||
+       spr->base.target == PIPE_TEXTURE_CUBE_ARRAY ||
+       spr->base.target == PIPE_TEXTURE_CUBE ||
+       spr->base.target == PIPE_TEXTURE_3D)
+      base_layer = r_coord + iview->u.tex.first_layer;
+   return softpipe_get_tex_image_offset(spr, iview->u.tex.level, base_layer);
+}
+
+/*
+ * Does this texture instruction have a layer or depth parameter.
+ */
+static inline bool
+has_layer_or_depth(unsigned tgsi_tex_instr)
+{
+   return (tgsi_tex_instr == TGSI_TEXTURE_3D ||
+           tgsi_tex_instr == TGSI_TEXTURE_CUBE ||
+           tgsi_tex_instr == TGSI_TEXTURE_1D_ARRAY ||
+           tgsi_tex_instr == TGSI_TEXTURE_2D_ARRAY ||
+           tgsi_tex_instr == TGSI_TEXTURE_CUBE_ARRAY ||
+           tgsi_tex_instr == TGSI_TEXTURE_2D_ARRAY_MSAA);
+}
+
+/*
+ * Is this texture instruction a single non-array coordinate.
+ */
+static inline bool
+has_1coord(unsigned tgsi_tex_instr)
+{
+   return (tgsi_tex_instr == TGSI_TEXTURE_BUFFER ||
+           tgsi_tex_instr == TGSI_TEXTURE_1D ||
+           tgsi_tex_instr == TGSI_TEXTURE_1D_ARRAY);
+}
+
+/*
+ * check the bounds vs w/h/d
+ */
+static inline bool
+bounds_check(int width, int height, int depth,
+             int s, int t, int r)
+{
+   if (s < 0 || s >= width)
+      return false;
+   if (t < 0 || t >= height)
+      return false;
+   if (r < 0 || r >= depth)
+      return false;
+   return true;
+}
+
+/*
+ * Checks if the texture target compatible with the image resource
+ * pipe target.
+ */
+static inline bool
+has_compat_target(unsigned pipe_target, unsigned tgsi_target)
+{
+   switch (pipe_target) {
+   case PIPE_TEXTURE_1D:
+      if (tgsi_target == TGSI_TEXTURE_1D)
+         return true;
+      break;
+   case PIPE_TEXTURE_2D:
+      if (tgsi_target == TGSI_TEXTURE_2D)
+         return true;
+      break;
+   case PIPE_TEXTURE_RECT:
+      if (tgsi_target == TGSI_TEXTURE_RECT)
+         return true;
+      break;
+   case PIPE_TEXTURE_3D:
+      if (tgsi_target == TGSI_TEXTURE_3D ||
+          tgsi_target == TGSI_TEXTURE_2D)
+         return true;
+      break;
+   case PIPE_TEXTURE_CUBE:
+      if (tgsi_target == TGSI_TEXTURE_CUBE ||
+          tgsi_target == TGSI_TEXTURE_2D)
+         return true;
+      break;
+   case PIPE_TEXTURE_1D_ARRAY:
+      if (tgsi_target == TGSI_TEXTURE_1D ||
+          tgsi_target == TGSI_TEXTURE_1D_ARRAY)
+         return true;
+      break;
+   case PIPE_TEXTURE_2D_ARRAY:
+      if (tgsi_target == TGSI_TEXTURE_2D ||
+          tgsi_target == TGSI_TEXTURE_2D_ARRAY)
+         return true;
+      break;
+   case PIPE_TEXTURE_CUBE_ARRAY:
+      if (tgsi_target == TGSI_TEXTURE_CUBE ||
+          tgsi_target == TGSI_TEXTURE_CUBE_ARRAY ||
+          tgsi_target == TGSI_TEXTURE_2D)
+         return true;
+      break;
+   case PIPE_BUFFER:
+      return (tgsi_target == TGSI_TEXTURE_BUFFER);
+   }
+   return false;
+}
+
+static bool
+get_dimensions(const struct pipe_image_view *iview,
+               const struct softpipe_resource *spr,
+               unsigned tgsi_tex_instr,
+               enum pipe_format pformat,
+               unsigned *width,
+               unsigned *height,
+               unsigned *depth)
+{
+   if (tgsi_tex_instr == TGSI_TEXTURE_BUFFER) {
+      *width = iview->u.buf.last_element - iview->u.buf.first_element + 1;
+      *height = 1;
+      *depth = 1;
+      /*
+       * Bounds check the buffer size from the view
+       * and the buffer size from the underlying buffer.
+       */
+      if (util_format_get_stride(pformat, *width) >
+          util_format_get_stride(spr->base.format, spr->base.width0))
+         return false;
+   } else {
+      unsigned level;
+
+      level = spr->base.target == PIPE_BUFFER ? 0 : iview->u.tex.level;
+      *width = u_minify(spr->base.width0, level);
+      *height = u_minify(spr->base.height0, level);
+
+      if (spr->base.target == TGSI_TEXTURE_3D)
+         *depth = u_minify(spr->base.depth0, level);
+      else
+         *depth = spr->base.array_size;
+
+      /* Make sure the resource and view have compatiable formats */
+      if (util_format_get_blocksize(pformat) >
+          util_format_get_blocksize(spr->base.format))
+         return false;
+   }
+   return true;
+}
+
+static void
+fill_coords(const struct tgsi_image_params *params,
+            unsigned index,
+            const int s[TGSI_QUAD_SIZE],
+            const int t[TGSI_QUAD_SIZE],
+            const int r[TGSI_QUAD_SIZE],
+            int *s_coord, int *t_coord, int *r_coord)
+{
+   *s_coord = s[index];
+   *t_coord = has_1coord(params->tgsi_tex_instr) ? 0 : t[index];
+   *r_coord = has_layer_or_depth(params->tgsi_tex_instr) ?
+      (params->tgsi_tex_instr == TGSI_TEXTURE_1D_ARRAY ? t[index] : r[index]) : 0;
+}
+/*
+ * Implement the image LOAD operation.
+ */
+static void
+sp_tgsi_load(const struct tgsi_image *image,
+             const struct tgsi_image_params *params,
+             const int s[TGSI_QUAD_SIZE],
+             const int t[TGSI_QUAD_SIZE],
+             const int r[TGSI_QUAD_SIZE],
+             const int sample[TGSI_QUAD_SIZE],
+             float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
+{
+   struct sp_tgsi_image *sp_img = (struct sp_tgsi_image *)image;
+   struct pipe_image_view *iview;
+   struct softpipe_resource *spr;
+   unsigned width, height, depth;
+   unsigned stride;
+   int c, j;
+   char *data_ptr;
+   unsigned offset = 0;
+
+   if (params->unit > PIPE_MAX_SHADER_IMAGES)
+      goto fail_write_all_zero;
+   iview = &sp_img->sp_iview[params->unit];
+   spr = (struct softpipe_resource *)iview->resource;
+   if (!spr)
+      goto fail_write_all_zero;
+
+   if (!has_compat_target(spr->base.target, params->tgsi_tex_instr))
+      goto fail_write_all_zero;
+
+   if (!get_dimensions(iview, spr, params->tgsi_tex_instr,
+                       params->format, &width, &height, &depth))
+      return;
+
+   stride = util_format_get_stride(params->format, width);
+
+   for (j = 0; j < TGSI_QUAD_SIZE; j++) {
+      int s_coord, t_coord, r_coord;
+      bool fill_zero = false;
+
+      if (!(params->execmask & (1 << j)))
+         fill_zero = true;
+
+      fill_coords(params, j, s, t, r, &s_coord, &t_coord, &r_coord);
+      if (!bounds_check(width, height, depth,
+                        s_coord, t_coord, r_coord))
+         fill_zero = true;
+
+      if (fill_zero) {
+         int nc = util_format_get_nr_components(params->format);
+         int ival = util_format_is_pure_integer(params->format);
+         for (c = 0; c < 4; c++) {
+            rgba[c][j] = 0;
+            if (c == 3 && nc < 4) {
+               if (ival)
+                  ((int32_t *)rgba[c])[j] = 1;
+               else
+                  rgba[c][j] = 1.0;
+            }
+         }
+         continue;
+      }
+      offset = get_image_offset(spr, iview, params->format, r_coord);
+      data_ptr = (char *)spr->data + offset;
+
+      if (util_format_is_pure_sint(params->format)) {
+         int32_t sdata[4];
+
+         util_format_read_4i(params->format,
+                             sdata, 0,
+                             data_ptr, stride,
+                             s_coord, t_coord, 1, 1);
+         for (c = 0; c < 4; c++)
+            ((int32_t *)rgba[c])[j] = sdata[c];
+      } else if (util_format_is_pure_uint(params->format)) {
+         uint32_t sdata[4];
+         util_format_read_4ui(params->format,
+                             sdata, 0,
+                             data_ptr, stride,
+                             s_coord, t_coord, 1, 1);
+         for (c = 0; c < 4; c++)
+            ((uint32_t *)rgba[c])[j] = sdata[c];
+      } else {
+         float sdata[4];
+         util_format_read_4f(params->format,
+                             sdata, 0,
+                             data_ptr, stride,
+                             s_coord, t_coord, 1, 1);
+         for (c = 0; c < 4; c++)
+            rgba[c][j] = sdata[c];
+      }
+   }
+   return;
+fail_write_all_zero:
+   for (j = 0; j < TGSI_QUAD_SIZE; j++) {
+      for (c = 0; c < 4; c++)
+         rgba[c][j] = 0;
+   }
+   return;
+}
+
+/*
+ * Implement the image STORE operation.
+ */
+static void
+sp_tgsi_store(const struct tgsi_image *image,
+              const struct tgsi_image_params *params,
+              const int s[TGSI_QUAD_SIZE],
+              const int t[TGSI_QUAD_SIZE],
+              const int r[TGSI_QUAD_SIZE],
+              const int sample[TGSI_QUAD_SIZE],
+              float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
+{
+   struct sp_tgsi_image *sp_img = (struct sp_tgsi_image *)image;
+   struct pipe_image_view *iview;
+   struct softpipe_resource *spr;
+   unsigned width, height, depth;
+   unsigned stride;
+   char *data_ptr;
+   int j, c;
+   unsigned offset = 0;
+   unsigned pformat = params->format;
+
+   if (params->unit > PIPE_MAX_SHADER_IMAGES)
+      return;
+   iview = &sp_img->sp_iview[params->unit];
+   spr = (struct softpipe_resource *)iview->resource;
+   if (!spr)
+      return;
+   if (!has_compat_target(spr->base.target, params->tgsi_tex_instr))
+      return;
+
+   if (params->format == PIPE_FORMAT_NONE)
+      pformat = spr->base.format;
+
+   if (!get_dimensions(iview, spr, params->tgsi_tex_instr,
+                       pformat, &width, &height, &depth))
+      return;
+
+   stride = util_format_get_stride(pformat, width);
+
+   for (j = 0; j < TGSI_QUAD_SIZE; j++) {
+      int s_coord, t_coord, r_coord;
+
+      if (!(params->execmask & (1 << j)))
+         continue;
+
+      fill_coords(params, j, s, t, r, &s_coord, &t_coord, &r_coord);
+      if (!bounds_check(width, height, depth,
+                        s_coord, t_coord, r_coord))
+         continue;
+
+      offset = get_image_offset(spr, iview, pformat, r_coord);
+      data_ptr = (char *)spr->data + offset;
+
+      if (util_format_is_pure_sint(pformat)) {
+         int32_t sdata[4];
+         for (c = 0; c < 4; c++)
+            sdata[c] = ((int32_t *)rgba[c])[j];
+         util_format_write_4i(pformat, sdata, 0, data_ptr, stride,
+                              s_coord, t_coord, 1, 1);
+      } else if (util_format_is_pure_uint(pformat)) {
+         uint32_t sdata[4];
+         for (c = 0; c < 4; c++)
+            sdata[c] = ((uint32_t *)rgba[c])[j];
+         util_format_write_4ui(pformat, sdata, 0, data_ptr, stride,
+                               s_coord, t_coord, 1, 1);
+      } else {
+         float sdata[4];
+         for (c = 0; c < 4; c++)
+            sdata[c] = rgba[c][j];
+         util_format_write_4f(pformat, sdata, 0, data_ptr, stride,
+                              s_coord, t_coord, 1, 1);
+      }
+   }
+}
+
+/*
+ * Implement atomic operations on unsigned integers.
+ */
+static void
+handle_op_uint(const struct pipe_image_view *iview,
+               const struct tgsi_image_params *params,
+               bool just_read,
+               char *data_ptr,
+               uint qi,
+               unsigned stride,
+               unsigned opcode,
+               int s,
+               int t,
+               float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE],
+               float rgba2[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
+{
+   uint c;
+   int nc = util_format_get_nr_components(params->format);
+   unsigned sdata[4];
+
+   util_format_read_4ui(params->format,
+                        sdata, 0,
+                        data_ptr, stride,
+                        s, t, 1, 1);
+
+   if (just_read) {
+      for (c = 0; c < nc; c++) {
+         ((uint32_t *)rgba[c])[qi] = sdata[c];
+      }
+      return;
+   }
+   switch (opcode) {
+   case TGSI_OPCODE_ATOMUADD:
+      for (c = 0; c < nc; c++) {
+         unsigned temp = sdata[c];
+         sdata[c] += ((uint32_t *)rgba[c])[qi];
+         ((uint32_t *)rgba[c])[qi] = temp;
+      }
+      break;
+   case TGSI_OPCODE_ATOMXCHG:
+      for (c = 0; c < nc; c++) {
+         unsigned temp = sdata[c];
+         sdata[c] = ((uint32_t *)rgba[c])[qi];
+         ((uint32_t *)rgba[c])[qi] = temp;
+      }
+      break;
+   case TGSI_OPCODE_ATOMCAS:
+      for (c = 0; c < nc; c++) {
+         unsigned dst_x = sdata[c];
+         unsigned cmp_x = ((uint32_t *)rgba[c])[qi];
+         unsigned src_x = ((uint32_t *)rgba2[c])[qi];
+         unsigned temp = sdata[c];
+         sdata[c] = (dst_x == cmp_x) ? src_x : dst_x;
+         ((uint32_t *)rgba[c])[qi] = temp;
+      }
+      break;
+   case TGSI_OPCODE_ATOMAND:
+      for (c = 0; c < nc; c++) {
+         unsigned temp = sdata[c];
+         sdata[c] &= ((uint32_t *)rgba[c])[qi];
+         ((uint32_t *)rgba[c])[qi] = temp;
+      }
+      break;
+   case TGSI_OPCODE_ATOMOR:
+      for (c = 0; c < nc; c++) {
+         unsigned temp = sdata[c];
+         sdata[c] |= ((uint32_t *)rgba[c])[qi];
+         ((uint32_t *)rgba[c])[qi] = temp;
+      }
+      break;
+   case TGSI_OPCODE_ATOMXOR:
+      for (c = 0; c < nc; c++) {
+         unsigned temp = sdata[c];
+         sdata[c] ^= ((uint32_t *)rgba[c])[qi];
+         ((uint32_t *)rgba[c])[qi] = temp;
+      }
+      break;
+   case TGSI_OPCODE_ATOMUMIN:
+      for (c = 0; c < nc; c++) {
+         unsigned dst_x = sdata[c];
+         unsigned src_x = ((uint32_t *)rgba[c])[qi];
+         sdata[c] = MIN2(dst_x, src_x);
+         ((uint32_t *)rgba[c])[qi] = dst_x;
+      }
+      break;
+   case TGSI_OPCODE_ATOMUMAX:
+      for (c = 0; c < nc; c++) {
+         unsigned dst_x = sdata[c];
+         unsigned src_x = ((uint32_t *)rgba[c])[qi];
+         sdata[c] = MAX2(dst_x, src_x);
+         ((uint32_t *)rgba[c])[qi] = dst_x;
+      }
+      break;
+   case TGSI_OPCODE_ATOMIMIN:
+      for (c = 0; c < nc; c++) {
+         int dst_x = sdata[c];
+         int src_x = ((uint32_t *)rgba[c])[qi];
+         sdata[c] = MIN2(dst_x, src_x);
+         ((uint32_t *)rgba[c])[qi] = dst_x;
+      }
+      break;
+   case TGSI_OPCODE_ATOMIMAX:
+      for (c = 0; c < nc; c++) {
+         int dst_x = sdata[c];
+         int src_x = ((uint32_t *)rgba[c])[qi];
+         sdata[c] = MAX2(dst_x, src_x);
+         ((uint32_t *)rgba[c])[qi] = dst_x;
+      }
+      break;
+   default:
+      assert(!"Unexpected TGSI opcode in sp_tgsi_op");
+      break;
+   }
+   util_format_write_4ui(params->format, sdata, 0, data_ptr, stride,
+                         s, t, 1, 1);
+}
+
+/*
+ * Implement atomic operations on signed integers.
+ */
+static void
+handle_op_int(const struct pipe_image_view *iview,
+              const struct tgsi_image_params *params,
+              bool just_read,
+              char *data_ptr,
+              uint qi,
+              unsigned stride,
+              unsigned opcode,
+              int s,
+              int t,
+              float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE],
+              float rgba2[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
+{
+   uint c;
+   int nc = util_format_get_nr_components(params->format);
+   int sdata[4];
+   util_format_read_4i(params->format,
+                       sdata, 0,
+                       data_ptr, stride,
+                       s, t, 1, 1);
+
+   if (just_read) {
+      for (c = 0; c < nc; c++) {
+         ((int32_t *)rgba[c])[qi] = sdata[c];
+      }
+      return;
+   }
+   switch (opcode) {
+   case TGSI_OPCODE_ATOMUADD:
+      for (c = 0; c < nc; c++) {
+         int temp = sdata[c];
+         sdata[c] += ((int32_t *)rgba[c])[qi];
+         ((int32_t *)rgba[c])[qi] = temp;
+      }
+      break;
+   case TGSI_OPCODE_ATOMXCHG:
+      for (c = 0; c < nc; c++) {
+         int temp = sdata[c];
+         sdata[c] = ((int32_t *)rgba[c])[qi];
+         ((int32_t *)rgba[c])[qi] = temp;
+      }
+      break;
+   case TGSI_OPCODE_ATOMCAS:
+      for (c = 0; c < nc; c++) {
+         int dst_x = sdata[c];
+         int cmp_x = ((int32_t *)rgba[c])[qi];
+         int src_x = ((int32_t *)rgba2[c])[qi];
+         int temp = sdata[c];
+         sdata[c] = (dst_x == cmp_x) ? src_x : dst_x;
+         ((int32_t *)rgba[c])[qi] = temp;
+      }
+      break;
+   case TGSI_OPCODE_ATOMAND:
+      for (c = 0; c < nc; c++) {
+         int temp = sdata[c];
+         sdata[c] &= ((int32_t *)rgba[c])[qi];
+         ((int32_t *)rgba[c])[qi] = temp;
+      }
+      break;
+   case TGSI_OPCODE_ATOMOR:
+      for (c = 0; c < nc; c++) {
+         int temp = sdata[c];
+         sdata[c] |= ((int32_t *)rgba[c])[qi];
+         ((int32_t *)rgba[c])[qi] = temp;
+      }
+      break;
+   case TGSI_OPCODE_ATOMXOR:
+      for (c = 0; c < nc; c++) {
+         int temp = sdata[c];
+         sdata[c] ^= ((int32_t *)rgba[c])[qi];
+         ((int32_t *)rgba[c])[qi] = temp;
+      }
+      break;
+   case TGSI_OPCODE_ATOMUMIN:
+      for (c = 0; c < nc; c++) {
+         int dst_x = sdata[c];
+         int src_x = ((int32_t *)rgba[c])[qi];
+         sdata[c] = MIN2(dst_x, src_x);
+         ((int32_t *)rgba[c])[qi] = dst_x;
+      }
+      break;
+   case TGSI_OPCODE_ATOMUMAX:
+      for (c = 0; c < nc; c++) {
+         int dst_x = sdata[c];
+         int src_x = ((int32_t *)rgba[c])[qi];
+         sdata[c] = MAX2(dst_x, src_x);
+         ((int32_t *)rgba[c])[qi] = dst_x;
+      }
+      break;
+   case TGSI_OPCODE_ATOMIMIN:
+      for (c = 0; c < nc; c++) {
+         int dst_x = sdata[c];
+         int src_x = ((int32_t *)rgba[c])[qi];
+         sdata[c] = MIN2(dst_x, src_x);
+         ((int32_t *)rgba[c])[qi] = dst_x;
+      }
+      break;
+   case TGSI_OPCODE_ATOMIMAX:
+      for (c = 0; c < nc; c++) {
+         int dst_x = sdata[c];
+         int src_x = ((int32_t *)rgba[c])[qi];
+         sdata[c] = MAX2(dst_x, src_x);
+         ((int32_t *)rgba[c])[qi] = dst_x;
+      }
+      break;
+   default:
+      assert(!"Unexpected TGSI opcode in sp_tgsi_op");
+      break;
+   }
+   util_format_write_4i(params->format, sdata, 0, data_ptr, stride,
+                        s, t, 1, 1);
+}
+
+/*
+ * Implement atomic image operations.
+ */
+static void
+sp_tgsi_op(const struct tgsi_image *image,
+           const struct tgsi_image_params *params,
+           unsigned opcode,
+           const int s[TGSI_QUAD_SIZE],
+           const int t[TGSI_QUAD_SIZE],
+           const int r[TGSI_QUAD_SIZE],
+           const int sample[TGSI_QUAD_SIZE],
+           float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE],
+           float rgba2[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
+{
+   struct sp_tgsi_image *sp_img = (struct sp_tgsi_image *)image;
+   struct pipe_image_view *iview;
+   struct softpipe_resource *spr;
+   unsigned width, height, depth;
+   unsigned stride;
+   int j, c;
+   unsigned offset;
+   char *data_ptr;
+
+   if (params->unit > PIPE_MAX_SHADER_IMAGES)
+      return;
+   iview = &sp_img->sp_iview[params->unit];
+   spr = (struct softpipe_resource *)iview->resource;
+   if (!spr)
+      goto fail_write_all_zero;
+   if (!has_compat_target(spr->base.target, params->tgsi_tex_instr))
+      goto fail_write_all_zero;
+
+   if (!get_dimensions(iview, spr, params->tgsi_tex_instr,
+                       params->format, &width, &height, &depth))
+      goto fail_write_all_zero;
+
+   stride = util_format_get_stride(spr->base.format, width);
+
+   for (j = 0; j < TGSI_QUAD_SIZE; j++) {
+      int s_coord, t_coord, r_coord;
+      bool just_read = false;
+
+      fill_coords(params, j, s, t, r, &s_coord, &t_coord, &r_coord);
+      if (!bounds_check(width, height, depth,
+                        s_coord, t_coord, r_coord)) {
+         int nc = util_format_get_nr_components(params->format);
+         int ival = util_format_is_pure_integer(params->format);
+         int c;
+         for (c = 0; c < 4; c++) {
+            rgba[c][j] = 0;
+            if (c == 3 && nc < 4) {
+               if (ival)
+                  ((int32_t *)rgba[c])[j] = 1;
+               else
+                  rgba[c][j] = 1.0;
+            }
+         }
+         continue;
+      }
+
+      /* just readback the value for atomic if execmask isn't set */
+      if (!(params->execmask & (1 << j))) {
+         just_read = true;
+      }
+
+      offset = get_image_offset(spr, iview, params->format, r_coord);
+      data_ptr = (char *)spr->data + offset;
+
+      /* we should see atomic operations on r32 formats */
+      if (util_format_is_pure_uint(params->format))
+         handle_op_uint(iview, params, just_read, data_ptr, j, stride,
+                        opcode, s_coord, t_coord, rgba, rgba2);
+      else if (util_format_is_pure_sint(params->format))
+         handle_op_int(iview, params, just_read, data_ptr, j, stride,
+                       opcode, s_coord, t_coord, rgba, rgba2);
+      else
+         assert(0);
+   }
+   return;
+fail_write_all_zero:
+   for (j = 0; j < TGSI_QUAD_SIZE; j++) {
+      for (c = 0; c < 4; c++)
+         rgba[c][j] = 0;
+   }
+   return;
+}
+
+static void
+sp_tgsi_get_dims(const struct tgsi_image *image,
+                 const struct tgsi_image_params *params,
+                 int dims[4])
+{
+   struct sp_tgsi_image *sp_img = (struct sp_tgsi_image *)image;
+   struct pipe_image_view *iview;
+   struct softpipe_resource *spr;
+   int level;
+
+   if (params->unit > PIPE_MAX_SHADER_IMAGES)
+      return;
+   iview = &sp_img->sp_iview[params->unit];
+   spr = (struct softpipe_resource *)iview->resource;
+   if (!spr)
+      return;
+
+   if (params->tgsi_tex_instr == TGSI_TEXTURE_BUFFER) {
+      dims[0] = iview->u.buf.last_element - iview->u.buf.first_element + 1;
+      dims[1] = dims[2] = dims[3] = 0;
+      return;
+   }
+
+   level = iview->u.tex.level;
+   dims[0] = u_minify(spr->base.width0, level);
+   switch (params->tgsi_tex_instr) {
+   case TGSI_TEXTURE_1D_ARRAY:
+      dims[1] = iview->u.tex.last_layer - iview->u.tex.first_layer + 1;
+      /* fallthrough */
+   case TGSI_TEXTURE_1D:
+      return;
+   case TGSI_TEXTURE_2D_ARRAY:
+      dims[2] = iview->u.tex.last_layer - iview->u.tex.first_layer + 1;
+      /* fallthrough */
+   case TGSI_TEXTURE_2D:
+   case TGSI_TEXTURE_CUBE:
+   case TGSI_TEXTURE_RECT:
+      dims[1] = u_minify(spr->base.height0, level);
+      return;
+   case TGSI_TEXTURE_3D:
+      dims[1] = u_minify(spr->base.height0, level);
+      dims[2] = u_minify(spr->base.depth0, level);
+      return;
+   case TGSI_TEXTURE_CUBE_ARRAY:
+      dims[1] = u_minify(spr->base.height0, level);
+      dims[2] = (iview->u.tex.last_layer - iview->u.tex.first_layer + 1) / 6;
+      break;
+   default:
+      assert(!"unexpected texture target in sp_get_dims()");
+      return;
+   }
+}
+
+struct sp_tgsi_image *
+sp_create_tgsi_image(void)
+{
+   struct sp_tgsi_image *img = CALLOC_STRUCT(sp_tgsi_image);
+   if (!img)
+      return NULL;
+
+   img->base.load = sp_tgsi_load;
+   img->base.store = sp_tgsi_store;
+   img->base.op = sp_tgsi_op;
+   img->base.get_dims = sp_tgsi_get_dims;
+   return img;
+};