tgsi/exec: implement restartable machine.

This lets us restart the machine at a PC value, and exits
the machine when we hit a barrier.

Compute shaders will then execute all the threads up to the
barrier, then restart the machines after the barrier once
all are done.

Signed-off-by: Dave Airlie <airlied@redhat.com>

2 files changed, 29 insertions, 17 deletions

diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index 7ec30fb0e04..d914ef5d6f0 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -4781,7 +4781,7 @@ micro_umsb(union tgsi_exec_channel *dst,
    dst->i[3] = util_last_bit(src->u[3]) - 1;
 }
 
-static void
+static bool
 exec_instruction(
    struct tgsi_exec_machine *mach,
    const struct tgsi_full_instruction *inst,
@@ -5100,7 +5100,7 @@ exec_instruction(
             mach->CondStackTop = 0;
             mach->LoopStackTop = 0;
             *pc = -1;
-            return;
+            return true;
          }
 
          assert(mach->CallStackTop > 0);
@@ -5749,10 +5749,12 @@ exec_instruction(
       break;
    case TGSI_OPCODE_BARRIER:
    case TGSI_OPCODE_MEMBAR:
+      return false;
       break;
    default:
       assert( 0 );
    }
+   return true;
 }
 
 static void
@@ -5796,13 +5798,16 @@ uint
 tgsi_exec_machine_run( struct tgsi_exec_machine *mach, int start_pc )
 {
    uint i;
-   int pc = 0;
 
-   tgsi_exec_machine_setup_masks(mach);
+   mach->pc = start_pc;
+
+   if (!start_pc) {
+      tgsi_exec_machine_setup_masks(mach);
 
-   /* execute declarations (interpolants) */
-   for (i = 0; i < mach->NumDeclarations; i++) {
-      exec_declaration( mach, mach->Declarations+i );
+      /* execute declarations (interpolants) */
+      for (i = 0; i < mach->NumDeclarations; i++) {
+         exec_declaration( mach, mach->Declarations+i );
+      }
    }
 
    {
@@ -5811,24 +5816,29 @@ tgsi_exec_machine_run( struct tgsi_exec_machine *mach, int start_pc )
       struct tgsi_exec_vector outputs[PIPE_MAX_ATTRIBS];
       uint inst = 1;
 
-      memset(mach->Temps, 0, sizeof(temps));
-      if (mach->Outputs)
-         memset(mach->Outputs, 0, sizeof(outputs));
-      memset(temps, 0, sizeof(temps));
-      memset(outputs, 0, sizeof(outputs));
+      if (!start_pc) {
+         memset(mach->Temps, 0, sizeof(temps));
+         if (mach->Outputs)
+            memset(mach->Outputs, 0, sizeof(outputs));
+         memset(temps, 0, sizeof(temps));
+         memset(outputs, 0, sizeof(outputs));
+      }
 #endif
 
       /* execute instructions, until pc is set to -1 */
-      while (pc != -1) {
-
+      while (mach->pc != -1) {
+         bool barrier_hit;
 #if DEBUG_EXECUTION
          uint i;
 
-         tgsi_dump_instruction(&mach->Instructions[pc], inst++);
+         tgsi_dump_instruction(&mach->Instructions[mach->pc], inst++);
 #endif
 
-         assert(pc < (int) mach->NumInstructions);
-         exec_instruction(mach, mach->Instructions + pc, &pc);
+         assert(mach->pc < (int) mach->NumInstructions);
+         barrier_hit = !exec_instruction(mach, mach->Instructions + mach->pc, &mach->pc);
+
+         if (barrier_hit && mach->ShaderType == PIPE_SHADER_COMPUTE)
+            return 0;
 
 #if DEBUG_EXECUTION
          for (i = 0; i < TGSI_EXEC_NUM_TEMPS + TGSI_EXEC_NUM_TEMP_EXTRAS; i++) {
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.h b/src/gallium/auxiliary/tgsi/tgsi_exec.h
index 564b3d5d5cc..9343d788db7 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.h
@@ -445,6 +445,8 @@ struct tgsi_exec_machine
       SamplerViews[PIPE_MAX_SHADER_SAMPLER_VIEWS];
 
    boolean UsedGeometryShader;
+
+   int pc;
 };
 
 struct tgsi_exec_machine *