softpipe: add SSBO/shader atomics support.

This adds support for the features requires for ARB_shader_storage_buffer_object
and ARB_shader_atomic_counters, ARB_shader_atomic_counter_ops.

[airlied: some cleanups applied]
Reviewed-by: Brian Paul <[email protected]>
Reviewed-by: Roland Scheidegger <[email protected]>
Signed-off-by: Dave Airlie <[email protected]>

1 files changed, 371 insertions, 0 deletions

diff --git a/src/gallium/drivers/softpipe/sp_buffer.c b/src/gallium/drivers/softpipe/sp_buffer.c
new file mode 100644
index 00000000000..69717baa1a0
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_buffer.c
@@ -0,0 +1,371 @@
+/*
+ * 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_buffer.h"
+#include "sp_texture.h"
+
+#include "util/u_format.h"
+
+static bool
+get_dimensions(const struct pipe_shader_buffer *bview,
+               const struct softpipe_resource *spr,
+               unsigned *width)
+{
+   *width = bview->buffer_size;
+   /*
+    * Bounds check the buffer size from the view
+    * and the buffer size from the underlying buffer.
+    */
+   if (*width > spr->base.width0)
+      return false;
+   return true;
+}
+
+/*
+ * Implement the image LOAD operation.
+ */
+static void
+sp_tgsi_load(const struct tgsi_buffer *buffer,
+             const struct tgsi_buffer_params *params,
+             const int s[TGSI_QUAD_SIZE],
+             float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
+{
+   struct sp_tgsi_buffer *sp_buf = (struct sp_tgsi_buffer *)buffer;
+   struct pipe_shader_buffer *bview;
+   struct softpipe_resource *spr;
+   unsigned width;
+   int c, j;
+   unsigned char *data_ptr;
+   const struct util_format_description *format_desc = util_format_description(PIPE_FORMAT_R32_UINT);
+
+   if (params->unit > PIPE_MAX_SHADER_BUFFERS)
+      goto fail_write_all_zero;
+
+   bview = &sp_buf->sp_bview[params->unit];
+   spr = softpipe_resource(bview->buffer);
+   if (!spr)
+      goto fail_write_all_zero;
+
+   if (!get_dimensions(bview, spr, &width))
+      return;
+
+   for (j = 0; j < TGSI_QUAD_SIZE; j++) {
+      int s_coord;
+      bool fill_zero = false;
+      uint32_t sdata[4];
+
+      if (!(params->execmask & (1 << j)))
+         fill_zero = true;
+
+      s_coord = s[j];
+      if (s_coord >= width)
+         fill_zero = true;
+
+      if (fill_zero) {
+         for (c = 0; c < 4; c++)
+            rgba[c][j] = 0;
+         continue;
+      }
+      data_ptr = (unsigned char *)spr->data + bview->buffer_offset + s_coord;
+      for (c = 0; c < 4; c++) {
+         format_desc->fetch_rgba_uint(sdata, data_ptr, 0, 0);
+         ((uint32_t *)rgba[c])[j] = sdata[0];
+         data_ptr += 4;
+      }
+   }
+   return;
+fail_write_all_zero:
+   memset(rgba, 0, TGSI_NUM_CHANNELS * TGSI_QUAD_SIZE * 4);
+   return;
+}
+
+/*
+ * Implement the buffer STORE operation.
+ */
+static void
+sp_tgsi_store(const struct tgsi_buffer *buffer,
+              const struct tgsi_buffer_params *params,
+              const int s[TGSI_QUAD_SIZE],
+              float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
+{
+   struct sp_tgsi_buffer *sp_buf = (struct sp_tgsi_buffer *)buffer;
+   struct pipe_shader_buffer *bview;
+   struct softpipe_resource *spr;
+   unsigned width;
+   unsigned char *data_ptr;
+   int j, c;
+   const struct util_format_description *format_desc = util_format_description(PIPE_FORMAT_R32_UINT);
+
+   if (params->unit > PIPE_MAX_SHADER_BUFFERS)
+      return;
+
+   bview = &sp_buf->sp_bview[params->unit];
+   spr = softpipe_resource(bview->buffer);
+   if (!spr)
+      return;
+
+   if (!get_dimensions(bview, spr, &width))
+      return;
+
+   for (j = 0; j < TGSI_QUAD_SIZE; j++) {
+      int s_coord;
+
+      if (!(params->execmask & (1 << j)))
+         continue;
+
+      s_coord = s[j];
+      if (s_coord >= width)
+         continue;
+
+      data_ptr = (unsigned char *)spr->data + bview->buffer_offset + s_coord;
+
+      for (c = 0; c < 4; c++) {
+         if (params->writemask & (1 << c)) {
+            unsigned temp[4];
+            unsigned char *dptr = data_ptr + (c * 4);
+            temp[0] = ((uint32_t *)rgba[c])[j];
+            format_desc->pack_rgba_uint(dptr, 0, temp, 0, 1, 1);
+         }
+      }
+   }
+}
+
+/*
+ * Implement atomic operations on unsigned integers.
+ */
+static void
+handle_op_uint(const struct pipe_shader_buffer *bview,
+               bool just_read,
+               unsigned char *data_ptr,
+               uint qi,
+               unsigned opcode,
+               unsigned writemask,
+               float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE],
+               float rgba2[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
+{
+   uint c;
+   const struct util_format_description *format_desc = util_format_description(PIPE_FORMAT_R32_UINT);
+   unsigned sdata[4];
+
+   for (c = 0; c < 4; c++) {
+      unsigned temp[4];
+      unsigned char *dptr = data_ptr + (c * 4);
+      format_desc->fetch_rgba_uint(temp, dptr, 0, 0);
+      sdata[c] = temp[0];
+   }
+
+   if (just_read) {
+      for (c = 0; c < 4; c++) {
+         ((uint32_t *)rgba[c])[qi] = sdata[c];
+      }
+      return;
+   }
+
+   switch (opcode) {
+   case TGSI_OPCODE_ATOMUADD:
+      for (c = 0; c < 4; 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 < 4; 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 < 4; 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 < 4; 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 < 4; 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 < 4; 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 < 4; 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 < 4; 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 < 4; 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 < 4; 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;
+   }
+
+   for (c = 0; c < 4; c++) {
+      if (writemask & (1 << c)) {
+         unsigned temp[4];
+         unsigned char *dptr = data_ptr + (c * 4);
+         temp[0] = sdata[c];
+         format_desc->pack_rgba_uint(dptr, 0, temp, 0, 1, 1);
+      }
+   }
+}
+
+/*
+ * Implement atomic buffer operations.
+ */
+static void
+sp_tgsi_op(const struct tgsi_buffer *buffer,
+           const struct tgsi_buffer_params *params,
+           unsigned opcode,
+           const int s[TGSI_QUAD_SIZE],
+           float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE],
+           float rgba2[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
+{
+   struct sp_tgsi_buffer *sp_buf = (struct sp_tgsi_buffer *)buffer;
+   struct pipe_shader_buffer *bview;
+   struct softpipe_resource *spr;
+   unsigned width;
+   int j, c;
+   unsigned char *data_ptr;
+
+   if (params->unit > PIPE_MAX_SHADER_BUFFERS)
+      return;
+
+   bview = &sp_buf->sp_bview[params->unit];
+   spr = softpipe_resource(bview->buffer);
+   if (!spr)
+      goto fail_write_all_zero;
+
+   if (!get_dimensions(bview, spr, &width))
+      goto fail_write_all_zero;
+
+   for (j = 0; j < TGSI_QUAD_SIZE; j++) {
+      int s_coord;
+      bool just_read = false;
+
+      s_coord = s[j];
+      if (s_coord >= width) {
+         for (c = 0; c < 4; c++) {
+            rgba[c][j] = 0;
+         }
+         continue;
+      }
+
+      /* just readback the value for atomic if execmask isn't set */
+      if (!(params->execmask & (1 << j))) {
+         just_read = true;
+      }
+
+      data_ptr = (unsigned char *)spr->data + bview->buffer_offset + s_coord;
+      /* we should see atomic operations on r32 formats */
+
+      handle_op_uint(bview, just_read, data_ptr, j,
+                     opcode, params->writemask, rgba, rgba2);
+   }
+   return;
+fail_write_all_zero:
+   memset(rgba, 0, TGSI_NUM_CHANNELS * TGSI_QUAD_SIZE * 4);
+   return;
+}
+
+/*
+ * return size of the attached buffer for RESQ opcode.
+ */
+static void
+sp_tgsi_get_dims(const struct tgsi_buffer *buffer,
+                 const struct tgsi_buffer_params *params,
+                 int *dim)
+{
+   struct sp_tgsi_buffer *sp_buf = (struct sp_tgsi_buffer *)buffer;
+   struct pipe_shader_buffer *bview;
+   struct softpipe_resource *spr;
+
+   if (params->unit > PIPE_MAX_SHADER_BUFFERS)
+      return;
+
+   bview = &sp_buf->sp_bview[params->unit];
+   spr = softpipe_resource(bview->buffer);
+   if (!spr)
+      return;
+
+   *dim = bview->buffer_size;
+}
+
+struct sp_tgsi_buffer *
+sp_create_tgsi_buffer(void)
+{
+   struct sp_tgsi_buffer *buf = CALLOC_STRUCT(sp_tgsi_buffer);
+   if (!buf)
+      return NULL;
+
+   buf->base.load = sp_tgsi_load;
+   buf->base.store = sp_tgsi_store;
+   buf->base.op = sp_tgsi_op;
+   buf->base.get_dims = sp_tgsi_get_dims;
+   return buf;
+};