gallium: Add SWITCH, CASE, DEFAULT and ENDSWITCH opcodes to TGSI.

Provide reference implementation of them in tgsi_exec.

Note that BREAK opcode is overloaded and can be used to break out
of either a loop or a switch-case statement.

3 files changed, 179 insertions, 11 deletions

diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index e2f69850c6..0a276483c4 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -123,7 +123,7 @@
 
 /** The execution mask depends on the conditional mask and the loop mask */
 #define UPDATE_EXEC_MASK(MACH) \
-      MACH->ExecMask = MACH->CondMask & MACH->LoopMask & MACH->ContMask & MACH->FuncMask
+      MACH->ExecMask = MACH->CondMask & MACH->LoopMask & MACH->ContMask & MACH->Switch.mask & MACH->FuncMask
 
 
 static const union tgsi_exec_channel ZeroVec =
@@ -1796,6 +1796,90 @@ exec_vector_trinary(struct tgsi_exec_machine *mach,
 }
 
 static void
+exec_break(struct tgsi_exec_machine *mach)
+{
+   if (mach->BreakType == TGSI_EXEC_BREAK_INSIDE_LOOP) {
+      /* turn off loop channels for each enabled exec channel */
+      mach->LoopMask &= ~mach->ExecMask;
+      /* Todo: if mach->LoopMask == 0, jump to end of loop */
+      UPDATE_EXEC_MASK(mach);
+   } else {
+      assert(mach->BreakType == TGSI_EXEC_BREAK_INSIDE_SWITCH);
+
+      mach->Switch.mask = 0x0;
+
+      UPDATE_EXEC_MASK(mach);
+   }
+}
+
+static void
+exec_switch(struct tgsi_exec_machine *mach,
+            const struct tgsi_full_instruction *inst)
+{
+   assert(mach->SwitchStackTop < TGSI_EXEC_MAX_SWITCH_NESTING);
+   assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK);
+
+   mach->SwitchStack[mach->SwitchStackTop++] = mach->Switch;
+   fetch_source(mach, &mach->Switch.selector, &inst->Src[0], CHAN_X);
+   mach->Switch.mask = 0x0;
+   mach->Switch.defaultMask = 0x0;
+
+   mach->BreakStack[mach->BreakStackTop++] = mach->BreakType;
+   mach->BreakType = TGSI_EXEC_BREAK_INSIDE_SWITCH;
+
+   UPDATE_EXEC_MASK(mach);
+}
+
+static void
+exec_case(struct tgsi_exec_machine *mach,
+          const struct tgsi_full_instruction *inst)
+{
+   uint prevMask = mach->SwitchStack[mach->SwitchStackTop - 1].mask;
+   union tgsi_exec_channel src;
+   uint mask = 0;
+
+   fetch_source(mach, &src, &inst->Src[0], CHAN_X);
+
+   if (mach->Switch.selector.u[0] == src.u[0]) {
+      mask |= 0x1;
+   }
+   if (mach->Switch.selector.u[1] == src.u[1]) {
+      mask |= 0x2;
+   }
+   if (mach->Switch.selector.u[2] == src.u[2]) {
+      mask |= 0x4;
+   }
+   if (mach->Switch.selector.u[3] == src.u[3]) {
+      mask |= 0x8;
+   }
+
+   mach->Switch.defaultMask |= mask;
+
+   mach->Switch.mask |= mask & prevMask;
+
+   UPDATE_EXEC_MASK(mach);
+}
+
+static void
+exec_default(struct tgsi_exec_machine *mach)
+{
+   uint prevMask = mach->SwitchStack[mach->SwitchStackTop - 1].mask;
+
+   mach->Switch.mask |= ~mach->Switch.defaultMask & prevMask;
+
+   UPDATE_EXEC_MASK(mach);
+}
+
+static void
+exec_endswitch(struct tgsi_exec_machine *mach)
+{
+   mach->Switch = mach->SwitchStack[--mach->SwitchStackTop];
+   mach->BreakType = mach->BreakStack[--mach->BreakStackTop];
+
+   UPDATE_EXEC_MASK(mach);
+}
+
+static void
 micro_i2f(union tgsi_exec_channel *dst,
           const union tgsi_exec_channel *src)
 {
@@ -2841,6 +2925,8 @@ exec_instruction(
          mach->CallStack[mach->CallStackTop].CondStackTop = mach->CondStackTop;
          mach->CallStack[mach->CallStackTop].LoopStackTop = mach->LoopStackTop;
          mach->CallStack[mach->CallStackTop].ContStackTop = mach->ContStackTop;
+         mach->CallStack[mach->CallStackTop].SwitchStackTop = mach->SwitchStackTop;
+         mach->CallStack[mach->CallStackTop].BreakStackTop = mach->BreakStackTop;
          /* note that PC was already incremented above */
          mach->CallStack[mach->CallStackTop].ReturnAddr = *pc;
 
@@ -2848,12 +2934,17 @@ exec_instruction(
 
          /* Second, push the Cond, Loop, Cont, Func stacks */
          assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING);
-         mach->CondStack[mach->CondStackTop++] = mach->CondMask;
          assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
-         mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask;
          assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
-         mach->ContStack[mach->ContStackTop++] = mach->ContMask;
+         assert(mach->SwitchStackTop < TGSI_EXEC_MAX_SWITCH_NESTING);
+         assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK);
          assert(mach->FuncStackTop < TGSI_EXEC_MAX_CALL_NESTING);
+
+         mach->CondStack[mach->CondStackTop++] = mach->CondMask;
+         mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask;
+         mach->ContStack[mach->ContStackTop++] = mach->ContMask;
+         mach->SwitchStack[mach->SwitchStackTop++] = mach->Switch;
+         mach->BreakStack[mach->BreakStackTop++] = mach->BreakType;
          mach->FuncStack[mach->FuncStackTop++] = mach->FuncMask;
 
          /* Finally, jump to the subroutine */
@@ -2886,6 +2977,12 @@ exec_instruction(
          mach->ContStackTop = mach->CallStack[mach->CallStackTop].ContStackTop;
          mach->ContMask = mach->ContStack[mach->ContStackTop];
 
+         mach->SwitchStackTop = mach->CallStack[mach->CallStackTop].SwitchStackTop;
+         mach->Switch = mach->SwitchStack[mach->SwitchStackTop];
+
+         mach->BreakStackTop = mach->CallStack[mach->CallStackTop].BreakStackTop;
+         mach->BreakType = mach->BreakStack[mach->BreakStackTop];
+
          assert(mach->FuncStackTop > 0);
          mach->FuncMask = mach->FuncStack[--mach->FuncStackTop];
 
@@ -3180,11 +3277,15 @@ exec_instruction(
    case TGSI_OPCODE_BGNLOOP:
       /* push LoopMask and ContMasks */
       assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
-      mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask;
       assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
-      mach->ContStack[mach->ContStackTop++] = mach->ContMask;
       assert(mach->LoopLabelStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
+      assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK);
+
+      mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask;
+      mach->ContStack[mach->ContStackTop++] = mach->ContMask;
       mach->LoopLabelStack[mach->LoopLabelStackTop++] = *pc - 1;
+      mach->BreakStack[mach->BreakStackTop++] = mach->BreakType;
+      mach->BreakType = TGSI_EXEC_BREAK_INSIDE_LOOP;
       break;
 
    case TGSI_OPCODE_ENDFOR:
@@ -3231,6 +3332,8 @@ exec_instruction(
          --mach->LoopLabelStackTop;
          assert(mach->LoopCounterStackTop > 0);
          --mach->LoopCounterStackTop;
+
+         mach->BreakType = mach->BreakStack[--mach->BreakStackTop];
       }
       UPDATE_EXEC_MASK(mach);
       break;
@@ -3254,15 +3357,14 @@ exec_instruction(
          mach->ContMask = mach->ContStack[--mach->ContStackTop];
          assert(mach->LoopLabelStackTop > 0);
          --mach->LoopLabelStackTop;
+
+         mach->BreakType = mach->BreakStack[--mach->BreakStackTop];
       }
       UPDATE_EXEC_MASK(mach);
       break;
 
    case TGSI_OPCODE_BRK:
-      /* turn off loop channels for each enabled exec channel */
-      mach->LoopMask &= ~mach->ExecMask;
-      /* Todo: if mach->LoopMask == 0, jump to end of loop */
-      UPDATE_EXEC_MASK(mach);
+      exec_break(mach);
       break;
 
    case TGSI_OPCODE_CONT:
@@ -3293,6 +3395,12 @@ exec_instruction(
       mach->ContStackTop = mach->CallStack[mach->CallStackTop].ContStackTop;
       mach->ContMask = mach->ContStack[mach->ContStackTop];
 
+      mach->SwitchStackTop = mach->CallStack[mach->CallStackTop].SwitchStackTop;
+      mach->Switch = mach->SwitchStack[mach->SwitchStackTop];
+
+      mach->BreakStackTop = mach->CallStack[mach->CallStackTop].BreakStackTop;
+      mach->BreakType = mach->BreakStack[mach->BreakStackTop];
+
       assert(mach->FuncStackTop > 0);
       mach->FuncMask = mach->FuncStack[--mach->FuncStackTop];
 
@@ -3407,6 +3515,22 @@ exec_instruction(
       exec_vector_binary(mach, inst, micro_usne);
       break;
 
+   case TGSI_OPCODE_SWITCH:
+      exec_switch(mach, inst);
+      break;
+
+   case TGSI_OPCODE_CASE:
+      exec_case(mach, inst);
+      break;
+
+   case TGSI_OPCODE_DEFAULT:
+      exec_default(mach);
+      break;
+
+   case TGSI_OPCODE_ENDSWITCH:
+      exec_endswitch(mach);
+      break;
+
    default:
       assert( 0 );
    }
@@ -3431,9 +3555,13 @@ tgsi_exec_machine_run( struct tgsi_exec_machine *mach )
    mach->FuncMask = 0xf;
    mach->ExecMask = 0xf;
 
+   mach->Switch.mask = 0xf;
+
    assert(mach->CondStackTop == 0);
    assert(mach->LoopStackTop == 0);
    assert(mach->ContStackTop == 0);
+   assert(mach->SwitchStackTop == 0);
+   assert(mach->BreakStackTop == 0);
    assert(mach->CallStackTop == 0);
 
    mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] = 0;
@@ -3534,6 +3662,8 @@ tgsi_exec_machine_run( struct tgsi_exec_machine *mach )
    assert(mach->CondStackTop == 0);
    assert(mach->LoopStackTop == 0);
    assert(mach->ContStackTop == 0);
+   assert(mach->SwitchStackTop == 0);
+   assert(mach->BreakStackTop == 0);
    assert(mach->CallStackTop == 0);
 
    return ~mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0];
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.h b/src/gallium/auxiliary/tgsi/tgsi_exec.h
index afaf5c39c4..aa3a98d7f1 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.h
@@ -179,6 +179,7 @@ struct tgsi_exec_labels
 
 #define TGSI_EXEC_MAX_COND_NESTING  32
 #define TGSI_EXEC_MAX_LOOP_NESTING  32
+#define TGSI_EXEC_MAX_SWITCH_NESTING 32
 #define TGSI_EXEC_MAX_CALL_NESTING  32
 
 /* The maximum number of input attributes per vertex. For 2D
@@ -206,9 +207,29 @@ struct tgsi_call_record
    uint CondStackTop;
    uint LoopStackTop;
    uint ContStackTop;
+   int SwitchStackTop;
+   int BreakStackTop;
    uint ReturnAddr;
 };
 
+
+/* Switch-case block state. */
+struct tgsi_switch_record {
+   uint mask;                          /**< execution mask */
+   union tgsi_exec_channel selector;   /**< a value case statements are compared to */
+   uint defaultMask;                   /**< non-execute mask for default case */
+};
+
+
+enum tgsi_break_type {
+   TGSI_EXEC_BREAK_INSIDE_LOOP,
+   TGSI_EXEC_BREAK_INSIDE_SWITCH
+};
+
+
+#define TGSI_EXEC_MAX_BREAK_STACK (TGSI_EXEC_MAX_LOOP_NESTING + TGSI_EXEC_MAX_SWITCH_NESTING)
+
+
 /**
  * Run-time virtual machine state for executing TGSI shader.
  */
@@ -251,6 +272,12 @@ struct tgsi_exec_machine
    uint FuncMask;  /**< For function calls */
    uint ExecMask;  /**< = CondMask & LoopMask */
 
+   /* Current switch-case state. */
+   struct tgsi_switch_record Switch;
+
+   /* Current break type. */
+   enum tgsi_break_type BreakType;
+
    /** Condition mask stack (for nested conditionals) */
    uint CondStack[TGSI_EXEC_MAX_COND_NESTING];
    int CondStackTop;
@@ -271,6 +298,13 @@ struct tgsi_exec_machine
    uint ContStack[TGSI_EXEC_MAX_LOOP_NESTING];
    int ContStackTop;
 
+   /** Switch case stack */
+   struct tgsi_switch_record SwitchStack[TGSI_EXEC_MAX_SWITCH_NESTING];
+   int SwitchStackTop;
+
+   enum tgsi_break_type BreakStack[TGSI_EXEC_MAX_BREAK_STACK];
+   int BreakStackTop;
+
    /** Function execution mask stack (for executing subroutine code) */
    uint FuncStack[TGSI_EXEC_MAX_CALL_NESTING];
    int FuncStackTop;
diff --git a/src/gallium/auxiliary/tgsi/tgsi_info.c b/src/gallium/auxiliary/tgsi/tgsi_info.c
index c7cdd163ec..8e0635677a 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_info.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_info.c
@@ -170,7 +170,11 @@ static const struct tgsi_opcode_info opcode_info[TGSI_OPCODE_LAST] =
    { 1, 2, 0, 0, 0, 0, "USGE", TGSI_OPCODE_USGE },
    { 1, 2, 0, 0, 0, 0, "USHR", TGSI_OPCODE_USHR },
    { 1, 2, 0, 0, 0, 0, "USLT", TGSI_OPCODE_USLT },
-   { 1, 2, 0, 0, 0, 0, "USNE", TGSI_OPCODE_USNE }
+   { 1, 2, 0, 0, 0, 0, "USNE", TGSI_OPCODE_USNE },
+   { 0, 1, 0, 0, 0, 0, "SWITCH", TGSI_OPCODE_SWITCH },
+   { 0, 1, 0, 0, 0, 0, "CASE", TGSI_OPCODE_CASE },
+   { 0, 0, 0, 0, 0, 0, "DEFAULT", TGSI_OPCODE_DEFAULT },
+   { 0, 0, 0, 0, 0, 0, "ENDSWITCH", TGSI_OPCODE_ENDSWITCH }
 };
 
 const struct tgsi_opcode_info *