From e558786a3ed52222c07f916e213b63dcba1890a2 Mon Sep 17 00:00:00 2001
From: Eric Anholt <eric@anholt.net>
Date: Wed, 30 Jun 2010 17:05:11 -0700
Subject: [PATCH] i965: Add support for OPCODE_SSG.

The old compiler didn't use SSG, and instead emitted SGT/SGT/SUB.  We
can do a little better for SSG than we do for the SGT series.
---
 src/mesa/drivers/dri/i965/brw_vs_emit.c  | 20 ++++++++++++++++++
 src/mesa/drivers/dri/i965/brw_wm.h       |  4 ++++
 src/mesa/drivers/dri/i965/brw_wm_emit.c  | 26 ++++++++++++++++++++++++
 src/mesa/drivers/dri/i965/brw_wm_glsl.c  |  3 +++
 src/mesa/drivers/dri/i965/brw_wm_pass1.c |  1 +
 5 files changed, 54 insertions(+)

diff --git a/src/mesa/drivers/dri/i965/brw_vs_emit.c b/src/mesa/drivers/dri/i965/brw_vs_emit.c
index 0b44deeb634..128987d78a6 100644
--- a/src/mesa/drivers/dri/i965/brw_vs_emit.c
+++ b/src/mesa/drivers/dri/i965/brw_vs_emit.c
@@ -485,6 +485,23 @@ static void emit_cmp( struct brw_compile *p,
    brw_set_predicate_control(p, BRW_PREDICATE_NONE);
 }
 
+static void emit_sign(struct brw_vs_compile *c,
+		      struct brw_reg dst,
+		      struct brw_reg arg0)
+{
+   struct brw_compile *p = &c->func;
+
+   brw_MOV(p, dst, brw_imm_f(0));
+
+   brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, arg0, brw_imm_f(0));
+   brw_MOV(p, dst, brw_imm_f(-1.0));
+   brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+
+   brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_G, arg0, brw_imm_f(0));
+   brw_MOV(p, dst, brw_imm_f(1.0));
+   brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+}
+
 static void emit_max( struct brw_compile *p, 
 		      struct brw_reg dst,
 		      struct brw_reg arg0,
@@ -1719,6 +1736,9 @@ void brw_vs_emit(struct brw_vs_compile *c )
       case OPCODE_SLE:
          unalias2(c, dst, args[0], args[1], emit_sle);
          break;
+      case OPCODE_SSG:
+         unalias1(c, dst, args[0], emit_sign);
+         break;
       case OPCODE_SUB:
 	 brw_ADD(p, dst, args[0], negate(args[1]));
 	 break;
diff --git a/src/mesa/drivers/dri/i965/brw_wm.h b/src/mesa/drivers/dri/i965/brw_wm.h
index 277b6de4425..938557ff369 100644
--- a/src/mesa/drivers/dri/i965/brw_wm.h
+++ b/src/mesa/drivers/dri/i965/brw_wm.h
@@ -425,6 +425,10 @@ void emit_sop(struct brw_compile *p,
 	      GLuint cond,
 	      const struct brw_reg *arg0,
 	      const struct brw_reg *arg1);
+void emit_sign(struct brw_compile *p,
+	       const struct brw_reg *dst,
+	       GLuint mask,
+	       const struct brw_reg *arg0);
 void emit_tex(struct brw_wm_compile *c,
 	      struct brw_reg *dst,
 	      GLuint dst_flags,
diff --git a/src/mesa/drivers/dri/i965/brw_wm_emit.c b/src/mesa/drivers/dri/i965/brw_wm_emit.c
index 323cfac8fa7..11f482bddf2 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_emit.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_emit.c
@@ -668,6 +668,28 @@ void emit_cmp(struct brw_compile *p,
    }
 }
 
+void emit_sign(struct brw_compile *p,
+	       const struct brw_reg *dst,
+	       GLuint mask,
+	       const struct brw_reg *arg0)
+{
+   GLuint i;
+
+   for (i = 0; i < 4; i++) {
+      if (mask & (1<<i)) {
+	 brw_MOV(p, dst[i], brw_imm_f(0.0));
+
+	 brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, arg0[i], brw_imm_f(0));
+	 brw_MOV(p, dst[i], brw_imm_f(-1.0));
+	 brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+
+	 brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_G, arg0[i], brw_imm_f(0));
+	 brw_MOV(p, dst[i], brw_imm_f(1.0));
+	 brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+      }
+   }
+}
+
 void emit_max(struct brw_compile *p,
 	      const struct brw_reg *dst,
 	      GLuint mask,
@@ -1673,6 +1695,10 @@ void brw_wm_emit( struct brw_wm_compile *c )
 	 emit_sne(p, dst, dst_flags, args[0], args[1]);
 	break;
 
+      case OPCODE_SSG:
+	 emit_sign(p, dst, dst_flags, args[0]);
+	 break;
+
       case OPCODE_LIT:
 	 emit_lit(c, dst, dst_flags, args[0]);
 	 break;
diff --git a/src/mesa/drivers/dri/i965/brw_wm_glsl.c b/src/mesa/drivers/dri/i965/brw_wm_glsl.c
index fe3c89b7212..e23ce7ad7a5 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_glsl.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_glsl.c
@@ -1971,6 +1971,9 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c)
 		emit_sop(p, dst, dst_flags,
 			 BRW_CONDITIONAL_NEQ, args[0], args[1]);
 		break;
+	    case OPCODE_SSG:
+		emit_sign(p, dst, dst_flags, args[0]);
+		break;
 	    case OPCODE_MUL:
 		emit_alu2(p, brw_MUL, dst, dst_flags, args[0], args[1]);
 		break;
diff --git a/src/mesa/drivers/dri/i965/brw_wm_pass1.c b/src/mesa/drivers/dri/i965/brw_wm_pass1.c
index b4493940292..8ee1f153b1f 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_pass1.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_pass1.c
@@ -158,6 +158,7 @@ void brw_wm_pass1( struct brw_wm_compile *c )
       case OPCODE_FLR:
       case OPCODE_FRC:
       case OPCODE_MOV:
+      case OPCODE_SSG:
       case OPCODE_SWZ:
       case OPCODE_TRUNC:
 	 read0 = writemask;
-- 
2.30.2