gallium: improve the pipe_stream_output_info struct (v2)

There are 3 changes:

1) stride is specified for each buffer, not just one, so that drivers don't
   have to derive it from the outputs

2) new per-output property dst_offset, which specifies the offset
   into the buffer in dwords where the output should be stored,
   so that drivers don't have to compute the offsets manually;
   this will also be useful for gl_SkipComponents
   from ARB_transform_feedback3

3) register_mask is removed, instead, there is start_component
   and num_components; register_mask with non-consecutive 1s
   doesn't make much sense (some hardware cannot do packing of components)

Christoph Bumiller: fixed nvc0.

v2: resolve merge conflicts in Draw and clean it up

12 files changed, 50 insertions, 140 deletions

diff --git a/src/gallium/auxiliary/draw/draw_pt_so_emit.c b/src/gallium/auxiliary/draw/draw_pt_so_emit.c
index 466f46abab3..ecf287f9128 100644
--- a/src/gallium/auxiliary/draw/draw_pt_so_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_so_emit.c
@@ -79,72 +79,6 @@ void draw_pt_so_emit_prepare(struct pt_so_emit *emit)
    draw_do_flush( draw, DRAW_FLUSH_BACKEND );
 }
 
-static boolean
-is_component_writable(unsigned mask,
-                      unsigned compo)
-{
-   switch (mask) {
-   case TGSI_WRITEMASK_NONE:
-      return FALSE;
-   case TGSI_WRITEMASK_X:
-      return compo == 0;
-   case TGSI_WRITEMASK_Y:
-      return compo == 1;
-   case TGSI_WRITEMASK_XY:
-      return compo == 0 || compo == 1;
-   case TGSI_WRITEMASK_Z:
-      return compo == 2;
-   case TGSI_WRITEMASK_XZ:
-      return compo == 0 || compo == 2;
-   case TGSI_WRITEMASK_YZ:
-      return compo == 1 || compo == 2;
-   case TGSI_WRITEMASK_XYZ:
-      return compo == 0 || compo == 1 || compo == 2;
-   case TGSI_WRITEMASK_W:
-      return compo == 3;
-   case TGSI_WRITEMASK_XW:
-      return compo == 0 || compo == 3;
-   case TGSI_WRITEMASK_YW:
-      return compo == 1 || compo == 3;
-   case TGSI_WRITEMASK_XYW:
-      return compo == 0 || compo == 1 || compo == 3;
-   case TGSI_WRITEMASK_ZW:
-      return compo == 2 || compo == 3;
-   case TGSI_WRITEMASK_XZW:
-      return compo == 0 || compo == 1 || compo == 3;
-   case TGSI_WRITEMASK_YZW:
-      return compo == 1 || compo == 2 || compo == 4;
-   case TGSI_WRITEMASK_XYZW:
-      return compo < 4;
-   default:
-      debug_assert(!"Unknown writemask in stream out");
-      return compo < 4;
-   }
-}
-
-static INLINE int mask_num_comps(int register_mask)
-{
-   int comps = 0;
-   switch (register_mask) {
-   case TGSI_WRITEMASK_XYZW:
-      comps = 4;
-      break;
-   case TGSI_WRITEMASK_XYZ:
-      comps = 3;
-      break;
-   case TGSI_WRITEMASK_XY:
-      comps = 2;
-      break;
-   case TGSI_WRITEMASK_X:
-      comps = 1;
-      break;
-   default:
-      assert(0);
-      break;
-   }
-   return comps;
-}
-
 static void so_emit_prim(struct pt_so_emit *so,
                          unsigned *indices,
                          unsigned num_vertices)
@@ -170,14 +104,14 @@ static void so_emit_prim(struct pt_so_emit *so,
    /* check have we space to emit prim first - if not don't do anything */
    for (i = 0; i < num_vertices; ++i) {
       for (slot = 0; slot < state->num_outputs; ++slot) {
-         unsigned writemask = state->output[slot].register_mask;
+         unsigned num_comps = state->output[slot].num_components;
          int ob = state->output[slot].output_buffer;
 
-         if ((buffer_total_bytes[ob] + mask_num_comps(writemask) * sizeof(float)) >
+         if ((buffer_total_bytes[ob] + num_comps * sizeof(float)) >
              draw->so.targets[ob]->target.buffer_size) {
             return;
          }
-         buffer_total_bytes[ob] += mask_num_comps(writemask) * sizeof(float);
+         buffer_total_bytes[ob] += num_comps * sizeof(float);
       }
    }
 
@@ -190,21 +124,16 @@ static void so_emit_prim(struct pt_so_emit *so,
 
       for (slot = 0; slot < state->num_outputs; ++slot) {
          unsigned idx = state->output[slot].register_index;
-         unsigned writemask = state->output[slot].register_mask;
-         unsigned written_compos = 0;
-         unsigned compo;
+         unsigned start_comp = state->output[slot].start_component;
+         unsigned num_comps = state->output[slot].num_components;
          int ob = state->output[slot].output_buffer;
 
          buffer = (float *)((char *)draw->so.targets[ob]->mapping +
                             draw->so.targets[ob]->target.buffer_offset +
                             draw->so.targets[ob]->internal_offset);
-         for (compo = 0; compo < 4; ++compo) {
-            if (is_component_writable(writemask, compo)) {
-               buffer[written_compos++] = input[idx][compo];
-            }
-         }
-         draw->so.targets[ob]->internal_offset += written_compos * sizeof(float);
-         total_written_compos += written_compos;
+         memcpy(buffer, &input[idx][start_comp], num_comps * sizeof(float));
+         draw->so.targets[ob]->internal_offset += num_comps * sizeof(float);
+         total_written_compos += num_comps;
       }
    }
    so->emitted_vertices += num_vertices;
diff --git a/src/gallium/auxiliary/util/u_blitter.c b/src/gallium/auxiliary/util/u_blitter.c
index 59940d9cbe7..6a32de619ef 100644
--- a/src/gallium/auxiliary/util/u_blitter.c
+++ b/src/gallium/auxiliary/util/u_blitter.c
@@ -263,8 +263,8 @@ struct blitter_context *util_blitter_create(struct pipe_context *pipe)
 
       memset(&so, 0, sizeof(so));
       so.num_outputs = 1;
-      so.output[0].register_mask = TGSI_WRITEMASK_XYZW;
-      so.stride = 4;
+      so.output[0].num_components = 4;
+      so.stride[0] = 4;
 
       ctx->vs_pos_only =
          util_make_vertex_passthrough_shader_with_so(pipe, 1, semantic_names,
diff --git a/src/gallium/auxiliary/util/u_dump_state.c b/src/gallium/auxiliary/util/u_dump_state.c
index c728bc4021c..c346a0ba5ca 100644
--- a/src/gallium/auxiliary/util/u_dump_state.c
+++ b/src/gallium/auxiliary/util/u_dump_state.c
@@ -444,13 +444,15 @@ util_dump_shader_state(FILE *stream, const struct pipe_shader_state *state)
    util_dump_member_begin(stream, "stream_output");
    util_dump_struct_begin(stream, "pipe_stream_output_info");
    util_dump_member(stream, uint, &state->stream_output, num_outputs);
-   util_dump_member(stream, uint, &state->stream_output, stride);
+   util_dump_array(stream, uint, state->stream_output.stride,
+                   Elements(state->stream_output.stride));
    util_dump_array_begin(stream);
    for(i = 0; i < state->stream_output.num_outputs; ++i) {
       util_dump_elem_begin(stream);
       util_dump_struct_begin(stream, ""); /* anonymous */
       util_dump_member(stream, uint, &state->stream_output.output[i], register_index);
-      util_dump_member(stream, uint, &state->stream_output.output[i], register_mask);
+      util_dump_member(stream, uint, &state->stream_output.output[i], start_component);
+      util_dump_member(stream, uint, &state->stream_output.output[i], num_components);
       util_dump_member(stream, uint, &state->stream_output.output[i], output_buffer);
       util_dump_struct_end(stream);
       util_dump_elem_end(stream);
diff --git a/src/gallium/drivers/llvmpipe/lp_state_so.c b/src/gallium/drivers/llvmpipe/lp_state_so.c
index 108f3aa4f39..ed2272d05ee 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_so.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_so.c
@@ -42,7 +42,7 @@ llvmpipe_create_stream_output_state(struct pipe_context *pipe,
 
    if (so) {
       so->base.num_outputs = templ->num_outputs;
-      so->base.stride = templ->stride;
+      memcpy(so->base.stride, templ->stride, sizeof(templ->stride));
       memcpy(so->base.output, templ->output,
              templ->num_outputs * sizeof(templ->output[0]));
    }
diff --git a/src/gallium/drivers/nvc0/nvc0_program.c b/src/gallium/drivers/nvc0/nvc0_program.c
index cff76fe67f3..60abc224398 100644
--- a/src/gallium/drivers/nvc0/nvc0_program.c
+++ b/src/gallium/drivers/nvc0/nvc0_program.c
@@ -503,20 +503,17 @@ nvc0_program_create_tfb_state(const struct nv50_ir_prog_info *info,
       tfb->varying_count[b] = 0;
 
       for (i = 0; i < pso->num_outputs; ++i) {
+         unsigned startc = pso->output[i].start_component;
          if (pso->output[i].output_buffer != b)
             continue;
-         for (c = 0; c < 4; ++c) {
-            if (!(pso->output[i].register_mask & (1 << c)))
-               continue;
+         for (c = 0; c < pso->output[i].num_components; ++c) {
             tfb->varying_count[b]++;
             tfb->varying_index[n++] =
-               info->out[pso->output[i].register_index].slot[c];
+               info->out[pso->output[i].register_index].slot[startc + c];
          }
       }
-      tfb->stride[b] = tfb->varying_count[b] * 4;
+      tfb->stride[b] = pso->stride[b] * 4;
    }
-   if (pso->stride)
-      tfb->stride[0] = pso->stride;
 
    return tfb;
 }
diff --git a/src/gallium/drivers/r600/r600.h b/src/gallium/drivers/r600/r600.h
index 4bfb5a980f1..baf09c1d8aa 100644
--- a/src/gallium/drivers/r600/r600.h
+++ b/src/gallium/drivers/r600/r600.h
@@ -196,7 +196,7 @@ struct r600_so_target {
 
 	/* The buffer where BUFFER_FILLED_SIZE is stored. */
 	struct r600_resource	*filled_size;
-	unsigned		stride;
+	unsigned		stride_in_dw;
 	unsigned		so_index;
 };
 
@@ -248,7 +248,7 @@ struct r600_context {
 	struct r600_so_target		*so_targets[PIPE_MAX_SO_BUFFERS];
 	boolean				streamout_start;
 	unsigned			streamout_append_bitmask;
-	unsigned			*vs_shader_so_strides;
+	unsigned			*vs_so_stride_in_dw;
 };
 
 struct r600_draw {
diff --git a/src/gallium/drivers/r600/r600_hw_context.c b/src/gallium/drivers/r600/r600_hw_context.c
index b0a28d98215..ac90011dbe7 100644
--- a/src/gallium/drivers/r600/r600_hw_context.c
+++ b/src/gallium/drivers/r600/r600_hw_context.c
@@ -2033,7 +2033,7 @@ static void r600_set_streamout_enable(struct r600_context *ctx, unsigned buffer_
 void r600_context_streamout_begin(struct r600_context *ctx)
 {
 	struct r600_so_target **t = ctx->so_targets;
-	unsigned *strides = ctx->vs_shader_so_strides;
+	unsigned *stride_in_dw = ctx->vs_so_stride_in_dw;
 	unsigned buffer_en, i, update_flags = 0;
 
 	buffer_en = (ctx->num_so_targets >= 1 && t[0] ? 1 : 0) |
@@ -2064,7 +2064,7 @@ void r600_context_streamout_begin(struct r600_context *ctx)
 
 	for (i = 0; i < ctx->num_so_targets; i++) {
 		if (t[i]) {
-			t[i]->stride = strides[i];
+			t[i]->stride_in_dw = stride_in_dw[i];
 			t[i]->so_index = i;
 
 			update_flags |= SURFACE_BASE_UPDATE_STRMOUT(i);
@@ -2074,7 +2074,7 @@ void r600_context_streamout_begin(struct r600_context *ctx)
 							16*i - R600_CONTEXT_REG_OFFSET) >> 2;
 			ctx->pm4[ctx->pm4_cdwords++] = (t[i]->b.buffer_offset +
 							t[i]->b.buffer_size) >> 2; /* BUFFER_SIZE (in DW) */
-			ctx->pm4[ctx->pm4_cdwords++] = strides[i] >> 2;		   /* VTX_STRIDE (in DW) */
+			ctx->pm4[ctx->pm4_cdwords++] = stride_in_dw[i];		   /* VTX_STRIDE (in DW) */
 			ctx->pm4[ctx->pm4_cdwords++] = 0;			   /* BUFFER_BASE */
 
 			ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0);
@@ -2186,7 +2186,7 @@ void r600_context_draw_opaque_count(struct r600_context *ctx, struct r600_so_tar
 
 	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_CONTEXT_REG, 1, 0);
 	ctx->pm4[ctx->pm4_cdwords++] = (R_028B30_VGT_STRMOUT_DRAW_OPAQUE_VERTEX_STRIDE - R600_CONTEXT_REG_OFFSET) >> 2;
-	ctx->pm4[ctx->pm4_cdwords++] = t->stride >> 2;
+	ctx->pm4[ctx->pm4_cdwords++] = t->stride_in_dw;
 
 	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_COPY_DW, 4, 0);
 	ctx->pm4[ctx->pm4_cdwords++] = COPY_DW_SRC_IS_MEM | COPY_DW_DST_IS_REG;
diff --git a/src/gallium/drivers/r600/r600_pipe.h b/src/gallium/drivers/r600/r600_pipe.h
index bd782438354..45bb4da8c2a 100644
--- a/src/gallium/drivers/r600/r600_pipe.h
+++ b/src/gallium/drivers/r600/r600_pipe.h
@@ -147,7 +147,6 @@ struct r600_pipe_shader {
 	struct tgsi_token		*tokens;
 	unsigned	sprite_coord_enable;
 	struct pipe_stream_output_info	so;
-	unsigned			so_strides[4];
 };
 
 struct r600_pipe_sampler_state {
diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index 6121a43c763..6fd058069c1 100644
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -124,12 +124,14 @@ int r600_pipe_shader_create(struct pipe_context *ctx, struct r600_pipe_shader *s
 			unsigned i;
 			fprintf(stderr, "STREAMOUT\n");
 			for (i = 0; i < shader->so.num_outputs; i++) {
+				unsigned mask = ((1 << shader->so.output[i].num_components) - 1) <<
+						shader->so.output[i].start_component;
 				fprintf(stderr, "  %i: MEM_STREAM0_BUF%i OUT[%i].%s%s%s%s\n", i,
 					shader->so.output[i].output_buffer, shader->so.output[i].register_index,
-				        shader->so.output[i].register_mask & 1 ? "x" : "_",
-				        (shader->so.output[i].register_mask >> 1) & 1 ? "y" : "_",
-				        (shader->so.output[i].register_mask >> 2) & 1 ? "z" : "_",
-				        (shader->so.output[i].register_mask >> 3) & 1 ? "w" : "_");
+				        mask & 1 ? "x" : "_",
+				        (mask >> 1) & 1 ? "y" : "_",
+				        (mask >> 2) & 1 ? "z" : "_",
+				        (mask >> 3) & 1 ? "w" : "_");
 			}
 		}
 	}
@@ -863,11 +865,8 @@ static int r600_shader_from_tgsi(struct r600_pipe_context * rctx, struct r600_pi
 
 	/* Add stream outputs. */
 	if (ctx.type == TGSI_PROCESSOR_VERTEX && so.num_outputs) {
-		unsigned buffer_offset[PIPE_MAX_SO_BUFFERS] = {0};
-
 		for (i = 0; i < so.num_outputs; i++) {
 			struct r600_bytecode_output output;
-			unsigned comps;
 
 			if (so.output[i].output_buffer >= 4) {
 				R600_ERR("exceeded the max number of stream output buffers, got: %d\n",
@@ -875,36 +874,21 @@ static int r600_shader_from_tgsi(struct r600_pipe_context * rctx, struct r600_pi
 				r = -EINVAL;
 				goto out_err;
 			}
-
-			switch (so.output[i].register_mask) {
-			case TGSI_WRITEMASK_XYZW:
-				comps = 4;
-				break;
-			case TGSI_WRITEMASK_XYZ:
-				comps = 3;
-				break;
-			case TGSI_WRITEMASK_XY:
-				comps = 2;
-				break;
-			case TGSI_WRITEMASK_X:
-				comps = 1;
-				break;
-			default:
-				R600_ERR("streamout: invalid register_mask, got: %x\n",
-					 so.output[i].register_mask);
-				r = -EINVAL;
-				goto out_err;
+			if (so.output[i].start_component) {
+			   R600_ERR("stream_output - start_component cannot be non-zero\n");
+			   r = -EINVAL;
+			   goto out_err;
 			}
 
 			memset(&output, 0, sizeof(struct r600_bytecode_output));
 			output.gpr = shader->output[so.output[i].register_index].gpr;
 			output.elem_size = 0;
-			output.array_base = buffer_offset[so.output[i].output_buffer];
+			output.array_base = so.output[i].dst_offset;
 			output.type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_WRITE;
 			output.burst_count = 1;
 			output.barrier = 1;
 			output.array_size = 0;
-			output.comp_mask = so.output[i].register_mask;
+			output.comp_mask = (1 << so.output[i].num_components) - 1;
 			if (ctx.bc->chip_class >= EVERGREEN) {
 				switch (so.output[i].output_buffer) {
 				case 0:
@@ -939,12 +923,6 @@ static int r600_shader_from_tgsi(struct r600_pipe_context * rctx, struct r600_pi
 			r = r600_bytecode_add_output(ctx.bc, &output);
 			if (r)
 				goto out_err;
-
-			buffer_offset[so.output[i].output_buffer] += comps;
-		}
-
-		for (i = 0; i < PIPE_MAX_SO_BUFFERS; i++) {
-			pipeshader->so_strides[i] = buffer_offset[i] * 4;
 		}
 	}
 
diff --git a/src/gallium/drivers/r600/r600_state_common.c b/src/gallium/drivers/r600/r600_state_common.c
index 034a560a7ec..ac9bd27df24 100644
--- a/src/gallium/drivers/r600/r600_state_common.c
+++ b/src/gallium/drivers/r600/r600_state_common.c
@@ -649,7 +649,7 @@ void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *dinfo)
 		}
 	}
 
-	rctx->ctx.vs_shader_so_strides = rctx->vs_shader->so_strides;
+	rctx->ctx.vs_so_stride_in_dw = rctx->vs_shader->so.stride;
 
 	mask = (1ULL << ((unsigned)rctx->framebuffer.nr_cbufs * 4)) - 1;
 
diff --git a/src/gallium/drivers/trace/tr_dump_state.c b/src/gallium/drivers/trace/tr_dump_state.c
index 8af084c2769..fb8683adba5 100644
--- a/src/gallium/drivers/trace/tr_dump_state.c
+++ b/src/gallium/drivers/trace/tr_dump_state.c
@@ -274,14 +274,16 @@ void trace_dump_shader_state(const struct pipe_shader_state *state)
    trace_dump_member_begin("stream_output");
    trace_dump_struct_begin("pipe_stream_output_info");
    trace_dump_member(uint, &state->stream_output, num_outputs);
-   trace_dump_member(uint, &state->stream_output, stride);
+   trace_dump_array(uint, state->stream_output.stride, PIPE_MAX_SO_BUFFERS);
    trace_dump_array_begin();
    for(i = 0; i < state->stream_output.num_outputs; ++i) {
       trace_dump_elem_begin();
       trace_dump_struct_begin(""); /* anonymous */
       trace_dump_member(uint, &state->stream_output.output[i], register_index);
-      trace_dump_member(uint, &state->stream_output.output[i], register_mask);
+      trace_dump_member(uint, &state->stream_output.output[i], start_component);
+      trace_dump_member(uint, &state->stream_output.output[i], num_components);
       trace_dump_member(uint, &state->stream_output.output[i], output_buffer);
+      trace_dump_member(uint, &state->stream_output.output[i], dst_offset);
       trace_dump_struct_end();
       trace_dump_elem_end();
    }
diff --git a/src/gallium/include/pipe/p_state.h b/src/gallium/include/pipe/p_state.h
index 024d544e3ef..f6486f01dce 100644
--- a/src/gallium/include/pipe/p_state.h
+++ b/src/gallium/include/pipe/p_state.h
@@ -190,16 +190,19 @@ struct pipe_clip_state
 struct pipe_stream_output_info
 {
    unsigned num_outputs;
-   /** stride for an entire vertex, only used if all output_buffers are 0 */
-   unsigned stride;
+   /** stride for an entire vertex for each buffer in dwords */
+   unsigned stride[PIPE_MAX_SO_BUFFERS];
+
    /**
     * Array of stream outputs, in the order they are to be written in.
     * Selected components are tightly packed into the output buffer.
     */
    struct {
-      unsigned register_index:8; /**< 0 to PIPE_MAX_SHADER_OUTPUTS */
-      unsigned register_mask:4;  /**< TGSI_WRITEMASK_x */
-      unsigned output_buffer:4;  /**< 0 to PIPE_MAX_SO_BUFFERS */
+      unsigned register_index:8;  /**< 0 to PIPE_MAX_SHADER_OUTPUTS */
+      unsigned start_component:2; /** 0 to 3 */
+      unsigned num_components:3;  /** 1 to 4 */
+      unsigned output_buffer:3;   /**< 0 to PIPE_MAX_SO_BUFFERS */
+      unsigned dst_offset:16;     /**< offset into the buffer in dwords */
    } output[PIPE_MAX_SHADER_OUTPUTS];
 };