r300/compiler: implement SSG opcode

4 files changed, 85 insertions, 1 deletions

diff --git a/src/gallium/drivers/r300/r300_tgsi_to_rc.c b/src/gallium/drivers/r300/r300_tgsi_to_rc.c
index dd697b9c374..53f07c1f02b 100644
--- a/src/gallium/drivers/r300/r300_tgsi_to_rc.c
+++ b/src/gallium/drivers/r300/r300_tgsi_to_rc.c
@@ -97,7 +97,7 @@ static unsigned translate_opcode(unsigned opcode)
      /* case TGSI_OPCODE_BRA: return RC_OPCODE_BRA; */
      /* case TGSI_OPCODE_CAL: return RC_OPCODE_CAL; */
      /* case TGSI_OPCODE_RET: return RC_OPCODE_RET; */
-     /* case TGSI_OPCODE_SSG: return RC_OPCODE_SSG; */
+        case TGSI_OPCODE_SSG: return RC_OPCODE_SSG;
         case TGSI_OPCODE_CMP: return RC_OPCODE_CMP;
         case TGSI_OPCODE_SCS: return RC_OPCODE_SCS;
         case TGSI_OPCODE_TXB: return RC_OPCODE_TXB;
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.c b/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.c
index 2ea830be7f9..0602e02a98b 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.c
@@ -295,6 +295,13 @@ struct rc_opcode_info rc_opcodes[MAX_RC_OPCODE] = {
 		.IsComponentwise = 1
 	},
 	{
+		.Opcode = RC_OPCODE_SSG,
+		.Name = "SSG",
+		.NumSrcRegs = 1,
+		.HasDstReg = 1,
+		.IsComponentwise = 1
+	},
+	{
 		.Opcode = RC_OPCODE_SUB,
 		.Name = "SUB",
 		.NumSrcRegs = 2,
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.h b/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.h
index 6e18d6eb3f1..0a9fceef386 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.h
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.h
@@ -154,6 +154,9 @@ typedef enum {
 	/** vec4 instruction: dst.c = (src0.c != src1.c) ? 1.0 : 0.0 */
 	RC_OPCODE_SNE,
 
+	/** vec4 instruction: dst.c = (src0.c < 0 ?) -1 : ((src0.c > 0) : 1 : 0) */
+	RC_OPCODE_SSG,
+
 	/** vec4 instruction: dst.c = src0.c - src1.c */
 	RC_OPCODE_SUB,
 
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c b/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c
index 857aae55145..c988e0c67e2 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c
@@ -464,6 +464,43 @@ static void transform_SNE(struct radeon_compiler* c,
 	rc_remove_instruction(inst);
 }
 
+static void transform_SSG(struct radeon_compiler* c,
+	struct rc_instruction* inst)
+{
+	/* result = sign(x)
+	 *
+	 *   CMP tmp0, -x, 1, 0
+	 *   CMP tmp1, x, 1, 0
+	 *   ADD result, tmp0, -tmp1;
+	 */
+	unsigned tmp0, tmp1;
+
+	/* 0 < x */
+	tmp0 = rc_find_free_temporary(c);
+	emit3(c, inst->Prev, RC_OPCODE_CMP, 0,
+	      dstregtmpmask(tmp0, inst->U.I.DstReg.WriteMask),
+	      negate(inst->U.I.SrcReg[0]),
+	      builtin_one,
+	      builtin_zero);
+
+	/* x < 0 */
+	tmp1 = rc_find_free_temporary(c);
+	emit3(c, inst->Prev, RC_OPCODE_CMP, 0,
+	      dstregtmpmask(tmp1, inst->U.I.DstReg.WriteMask),
+	      inst->U.I.SrcReg[0],
+	      builtin_one,
+	      builtin_zero);
+
+	/* Either both are zero, or one of them is one and the other is zero. */
+	/* result = tmp0 - tmp1 */
+	emit2(c, inst->Prev, RC_OPCODE_ADD, 0,
+	      inst->U.I.DstReg,
+	      srcreg(RC_FILE_TEMPORARY, tmp0),
+	      negate(srcreg(RC_FILE_TEMPORARY, tmp1)));
+
+	rc_remove_instruction(inst);
+}
+
 static void transform_SUB(struct radeon_compiler* c,
 	struct rc_instruction* inst)
 {
@@ -530,6 +567,7 @@ int radeonTransformALU(
 	case RC_OPCODE_SLE: transform_SLE(c, inst); return 1;
 	case RC_OPCODE_SLT: transform_SLT(c, inst); return 1;
 	case RC_OPCODE_SNE: transform_SNE(c, inst); return 1;
+	case RC_OPCODE_SSG: transform_SSG(c, inst); return 1;
 	case RC_OPCODE_SUB: transform_SUB(c, inst); return 1;
 	case RC_OPCODE_SWZ: transform_SWZ(c, inst); return 1;
 	case RC_OPCODE_XPD: transform_XPD(c, inst); return 1;
@@ -672,6 +710,41 @@ static void transform_r300_vertex_SLE(struct radeon_compiler* c,
 	inst->U.I.SrcReg[1].Negate ^= RC_MASK_XYZW;
 }
 
+static void transform_r300_vertex_SSG(struct radeon_compiler* c,
+	struct rc_instruction* inst)
+{
+	/* result = sign(x)
+	 *
+	 *   SLT tmp0, 0, x;
+	 *   SLT tmp1, x, 0;
+	 *   ADD result, tmp0, -tmp1;
+	 */
+	unsigned tmp0, tmp1;
+
+	/* 0 < x */
+	tmp0 = rc_find_free_temporary(c);
+	emit2(c, inst->Prev, RC_OPCODE_SLT, 0,
+	      dstregtmpmask(tmp0, inst->U.I.DstReg.WriteMask),
+	      builtin_zero,
+	      inst->U.I.SrcReg[0]);
+
+	/* x < 0 */
+	tmp1 = rc_find_free_temporary(c);
+	emit2(c, inst->Prev, RC_OPCODE_SLT, 0,
+	      dstregtmpmask(tmp1, inst->U.I.DstReg.WriteMask),
+	      inst->U.I.SrcReg[0],
+	      builtin_zero);
+
+	/* Either both are zero, or one of them is one and the other is zero. */
+	/* result = tmp0 - tmp1 */
+	emit2(c, inst->Prev, RC_OPCODE_ADD, 0,
+	      inst->U.I.DstReg,
+	      srcreg(RC_FILE_TEMPORARY, tmp0),
+	      negate(srcreg(RC_FILE_TEMPORARY, tmp1)));
+
+	rc_remove_instruction(inst);
+}
+
 /**
  * For use with radeonLocalTransform, this transforms non-native ALU
  * instructions of the r300 up to r500 vertex engine.
@@ -705,6 +778,7 @@ int r300_transform_vertex_alu(
 			return 1;
 		}
 		return 0;
+	case RC_OPCODE_SSG: transform_r300_vertex_SSG(c, inst); return 1;
 	case RC_OPCODE_SUB: transform_SUB(c, inst); return 1;
 	case RC_OPCODE_SWZ: transform_SWZ(c, inst); return 1;
 	case RC_OPCODE_XPD: transform_XPD(c, inst); return 1;