[Mesa-dev] [PATCH 6/6] gallivm: implement breakc and implicit primitive flushing

[Zack Rusin]

we were missing implementation of the breakc instruction and our
TGSI semantics currently require an implicit endprim at the end
of GS if none is present - this implements both.

Signed-off-by: Zack Rusin <zackr at vmware.com>
---
 src/gallium/auxiliary/gallivm/lp_bld_tgsi.h        |    6 ++++
 src/gallium/auxiliary/gallivm/lp_bld_tgsi_action.c |    1 +
 src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c    |   38 ++++++++++++++++++++
 3 files changed, 45 insertions(+)

diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h b/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h
index 4c6456e..4acc592 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h
@@ -392,6 +392,12 @@ struct lp_build_tgsi_soa_context
    LLVMValueRef emitted_prims_vec;
    LLVMValueRef total_emitted_vertices_vec;
    LLVMValueRef emitted_vertices_vec;
+   /* if a shader doesn't have ENDPRIM instruction but it has
+    * a number of EMIT instructions it means the END instruction
+    * implicitly invokes ENDPRIM. handle this via a flag here
+    * in the future maybe we can enforce TGSI to always have
+    * an explicit ENDPRIM */
+   boolean pending_end_primitive;
 
    LLVMValueRef consts_ptr;
    const LLVMValueRef *pos;
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_action.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_action.c
index 41ddd99..55bb8e3 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_action.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_action.c
@@ -867,6 +867,7 @@ lp_set_default_actions(struct lp_build_tgsi_context * bld_base)
    bld_base->op_actions[TGSI_OPCODE_COS].fetch_args = scalar_unary_fetch_args;
    bld_base->op_actions[TGSI_OPCODE_EX2].fetch_args = scalar_unary_fetch_args;
    bld_base->op_actions[TGSI_OPCODE_IF].fetch_args = scalar_unary_fetch_args;
+   bld_base->op_actions[TGSI_OPCODE_BREAKC].fetch_args = scalar_unary_fetch_args;
    bld_base->op_actions[TGSI_OPCODE_KIL].fetch_args = kil_fetch_args;
    bld_base->op_actions[TGSI_OPCODE_KILP].fetch_args = kilp_fetch_args;
    bld_base->op_actions[TGSI_OPCODE_RCP].fetch_args = scalar_unary_fetch_args;
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
index 95633ab..36e49ac 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
@@ -213,6 +213,23 @@ static void lp_exec_break(struct lp_exec_mask *mask)
    lp_exec_mask_update(mask);
 }
 
+
+static void lp_exec_break_condition(struct lp_exec_mask *mask, LLVMValueRef cond)
+{
+   LLVMBuilderRef builder = mask->bld->gallivm->builder;
+   LLVMValueRef exec_mask = LLVMBuildNot(builder,
+                                         mask->exec_mask,
+                                         "break");
+
+   exec_mask = LLVMBuildAnd(builder, exec_mask, cond, "");
+
+   mask->break_mask = LLVMBuildAnd(builder,
+                                   mask->break_mask,
+                                   exec_mask, "break_full");
+
+   lp_exec_mask_update(mask);
+}
+
 static void lp_exec_continue(struct lp_exec_mask *mask)
 {
    LLVMBuilderRef builder = mask->bld->gallivm->builder;
@@ -2190,6 +2207,7 @@ emit_vertex(
          LLVMBuildAdd(builder, bld->emitted_vertices_vec, masked_ones, "");
       bld->total_emitted_vertices_vec =
          LLVMBuildAdd(builder, bld->total_emitted_vertices_vec, masked_ones, "");
+      bld->pending_end_primitive = TRUE;
    }
 }
 
@@ -2212,6 +2230,7 @@ end_primitive(
       bld->emitted_prims_vec =
          LLVMBuildAdd(builder, bld->emitted_prims_vec, masked_ones, "");
       bld->emitted_vertices_vec = bld_base->uint_bld.zero;
+      bld->pending_end_primitive = FALSE;
    }
 }
 
@@ -2250,6 +2269,17 @@ brk_emit(
 }
 
 static void
+breakc_emit(
+   const struct lp_build_tgsi_action * action,
+   struct lp_build_tgsi_context * bld_base,
+   struct lp_build_emit_data * emit_data)
+{
+   struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
+
+   lp_exec_break_condition(&bld->exec_mask, emit_data->args[0]);
+}
+
+static void
 if_emit(
    const struct lp_build_tgsi_action * action,
    struct lp_build_tgsi_context * bld_base,
@@ -2504,6 +2534,12 @@ static void emit_epilogue(struct lp_build_tgsi_context * bld_base)
    /* If we have indirect addressing in outputs we need to copy our alloca array
     * to the outputs slots specified by the caller */
    if (bld->gs_args) {
+      /* flush the accumulated vertices as a primitive */
+      if (bld->pending_end_primitive) {
+         end_primitive(NULL, bld_base, NULL);
+         bld->pending_end_primitive = FALSE;
+      }
+      
       bld->gs_args->gs_epilogue(&bld->bld_base,
                                 bld->total_emitted_vertices_vec,
                                 bld->emitted_prims_vec,
@@ -2572,6 +2608,7 @@ lp_build_tgsi_soa(struct gallivm_state *gallivm,
    bld.bld_base.op_actions[TGSI_OPCODE_BGNLOOP].emit = bgnloop_emit;
    bld.bld_base.op_actions[TGSI_OPCODE_BGNSUB].emit = bgnsub_emit;
    bld.bld_base.op_actions[TGSI_OPCODE_BRK].emit = brk_emit;
+   bld.bld_base.op_actions[TGSI_OPCODE_BREAKC].emit = breakc_emit;
    bld.bld_base.op_actions[TGSI_OPCODE_CAL].emit = cal_emit;
    bld.bld_base.op_actions[TGSI_OPCODE_CONT].emit = cont_emit;
    bld.bld_base.op_actions[TGSI_OPCODE_DDX].emit = ddx_emit;
@@ -2607,6 +2644,7 @@ lp_build_tgsi_soa(struct gallivm_state *gallivm,
       /* inputs are always indirect with gs */
       bld.indirect_files |= (1 << TGSI_FILE_INPUT);
       bld.gs_args = gs_args;
+      bld.pending_end_primitive = FALSE;
       bld.bld_base.emit_fetch_funcs[TGSI_FILE_INPUT] = emit_fetch_gs_input;
       bld.bld_base.op_actions[TGSI_OPCODE_EMIT].emit = emit_vertex;
       bld.bld_base.op_actions[TGSI_OPCODE_ENDPRIM].emit = end_primitive;
-- 
1.7.10.4