glsl: Add ir_unop_sin_reduced and ir_unop_cos_reduced

The operate just like ir_unop_sin and ir_unop_cos except that they
expect their inputs to be limited to the range [-pi, pi].  Several
GPUs require this limited range for their sine and cosine
instructions, so having these as operations (along with a to-be-written
lowering pass) helps this architectures.

These new operations also matche the semantics of the
GL_ARB_fragment_program SCS instruction.  Having these as operations
helps in generating GLSL IR directly from assembly fragment programs.

6 files changed, 120 insertions, 0 deletions

diff --git a/src/glsl/ir.cpp b/src/glsl/ir.cpp
index 574ef3e183..714826343c 100644
--- a/src/glsl/ir.cpp
+++ b/src/glsl/ir.cpp
@@ -239,6 +239,8 @@ static const char *const operator_strs[] = {
    "round_even",
    "sin",
    "cos",
+   "sin_reduced",
+   "cos_reduced",
    "dFdx",
    "dFdy",
    "noise",
diff --git a/src/glsl/ir.h b/src/glsl/ir.h
index 3ea7301f47..2b94e63cc2 100644
--- a/src/glsl/ir.h
+++ b/src/glsl/ir.h
@@ -729,6 +729,8 @@ enum ir_expression_operation {
    /*@{*/
    ir_unop_sin,
    ir_unop_cos,
+   ir_unop_sin_reduced,    /**< Reduced range sin. [-pi, pi] */
+   ir_unop_cos_reduced,    /**< Reduced range cos. [-pi, pi] */
    /*@}*/
 
    /**
diff --git a/src/glsl/ir_constant_expression.cpp b/src/glsl/ir_constant_expression.cpp
index 8a54fc78cc..45860b279f 100644
--- a/src/glsl/ir_constant_expression.cpp
+++ b/src/glsl/ir_constant_expression.cpp
@@ -216,6 +216,7 @@ ir_expression::constant_expression_value()
       break;
 
    case ir_unop_sin:
+   case ir_unop_sin_reduced:
       assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
       for (unsigned c = 0; c < op[0]->type->components(); c++) {
 	 data.f[c] = sinf(op[0]->value.f[c]);
@@ -223,6 +224,7 @@ ir_expression::constant_expression_value()
       break;
 
    case ir_unop_cos:
+   case ir_unop_cos_reduced:
       assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
       for (unsigned c = 0; c < op[0]->type->components(); c++) {
 	 data.f[c] = cosf(op[0]->value.f[c]);
diff --git a/src/glsl/ir_validate.cpp b/src/glsl/ir_validate.cpp
index 77f48968b8..2a066c1a27 100644
--- a/src/glsl/ir_validate.cpp
+++ b/src/glsl/ir_validate.cpp
@@ -273,6 +273,8 @@ ir_validate::visit_leave(ir_expression *ir)
    case ir_unop_fract:
    case ir_unop_sin:
    case ir_unop_cos:
+   case ir_unop_sin_reduced:
+   case ir_unop_cos_reduced:
    case ir_unop_dFdx:
    case ir_unop_dFdy:
       assert(ir->operands[0]->type->base_type == GLSL_TYPE_FLOAT);
diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp b/src/mesa/drivers/dri/i965/brw_fs.cpp
index ac795e0bda..164f89eace 100644
--- a/src/mesa/drivers/dri/i965/brw_fs.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs.cpp
@@ -778,9 +778,11 @@ fs_visitor::visit(ir_expression *ir)
       assert(!"not reached: should be handled by ir_explog_to_explog2");
       break;
    case ir_unop_sin:
+   case ir_unop_sin_reduced:
       emit_math(FS_OPCODE_SIN, this->result, op[0]);
       break;
    case ir_unop_cos:
+   case ir_unop_cos_reduced:
       emit_math(FS_OPCODE_COS, this->result, op[0]);
       break;
 
diff --git a/src/mesa/program/ir_to_mesa.cpp b/src/mesa/program/ir_to_mesa.cpp
index 870fd6f25e..ef9f692f94 100644
--- a/src/mesa/program/ir_to_mesa.cpp
+++ b/src/mesa/program/ir_to_mesa.cpp
@@ -282,6 +282,10 @@ public:
 				   ir_to_mesa_src_reg src0,
 				   ir_to_mesa_src_reg src1);
 
+   void emit_scs(ir_instruction *ir, enum prog_opcode op,
+		 ir_to_mesa_dst_reg dst,
+		 const ir_to_mesa_src_reg &src);
+
    GLboolean try_emit_mad(ir_expression *ir,
 			  int mul_operand);
 
@@ -475,6 +479,10 @@ ir_to_mesa_visitor::ir_to_mesa_emit_scalar_op2(ir_instruction *ir,
       GLuint src0_swiz = GET_SWZ(src0.swizzle, i);
       GLuint src1_swiz = GET_SWZ(src1.swizzle, i);
       for (j = i + 1; j < 4; j++) {
+	 /* If there is another enabled component in the destination that is
+	  * derived from the same inputs, generate its value on this pass as
+	  * well.
+	  */
 	 if (!(done_mask & (1 << j)) &&
 	     GET_SWZ(src0.swizzle, j) == src0_swiz &&
 	     GET_SWZ(src1.swizzle, j) == src1_swiz) {
@@ -508,6 +516,102 @@ ir_to_mesa_visitor::ir_to_mesa_emit_scalar_op1(ir_instruction *ir,
    ir_to_mesa_emit_scalar_op2(ir, op, dst, src0, undef);
 }
 
+/**
+ * Emit an OPCODE_SCS instruction
+ *
+ * The \c SCS opcode functions a bit differently than the other Mesa (or
+ * ARB_fragment_program) opcodes.  Instead of splatting its result across all
+ * four components of the destination, it writes one value to the \c x
+ * component and another value to the \c y component.
+ *
+ * \param ir        IR instruction being processed
+ * \param op        Either \c OPCODE_SIN or \c OPCODE_COS depending on which
+ *                  value is desired.
+ * \param dst       Destination register
+ * \param src       Source register
+ */
+void
+ir_to_mesa_visitor::emit_scs(ir_instruction *ir, enum prog_opcode op,
+			     ir_to_mesa_dst_reg dst,
+			     const ir_to_mesa_src_reg &src)
+{
+   /* Vertex programs cannot use the SCS opcode.
+    */
+   if (this->prog->Target == GL_VERTEX_PROGRAM_ARB) {
+      ir_to_mesa_emit_scalar_op1(ir, op, dst, src);
+      return;
+   }
+
+   const unsigned component = (op == OPCODE_SIN) ? 0 : 1;
+   const unsigned scs_mask = (1U << component);
+   int done_mask = ~dst.writemask;
+   ir_to_mesa_src_reg tmp;
+
+   assert(op == OPCODE_SIN || op == OPCODE_COS);
+
+   /* If there are compnents in the destination that differ from the component
+    * that will be written by the SCS instrution, we'll need a temporary.
+    */
+   if (scs_mask != unsigned(dst.writemask)) {
+      tmp = get_temp(glsl_type::vec4_type);
+   }
+
+   for (unsigned i = 0; i < 4; i++) {
+      unsigned this_mask = (1U << i);
+      ir_to_mesa_src_reg src0 = src;
+
+      if ((done_mask & this_mask) != 0)
+	 continue;
+
+      /* The source swizzle specified which component of the source generates
+       * sine / cosine for the current component in the destination.  The SCS
+       * instruction requires that this value be swizzle to the X component.
+       * Replace the current swizzle with a swizzle that puts the source in
+       * the X component.
+       */
+      unsigned src0_swiz = GET_SWZ(src.swizzle, i);
+
+      src0.swizzle = MAKE_SWIZZLE4(src0_swiz, src0_swiz,
+				   src0_swiz, src0_swiz);
+      for (unsigned j = i + 1; j < 4; j++) {
+	 /* If there is another enabled component in the destination that is
+	  * derived from the same inputs, generate its value on this pass as
+	  * well.
+	  */
+	 if (!(done_mask & (1 << j)) &&
+	     GET_SWZ(src0.swizzle, j) == src0_swiz) {
+	    this_mask |= (1 << j);
+	 }
+      }
+
+      if (this_mask != scs_mask) {
+	 ir_to_mesa_instruction *inst;
+	 ir_to_mesa_dst_reg tmp_dst = ir_to_mesa_dst_reg_from_src(tmp);
+
+	 /* Emit the SCS instruction.
+	  */
+	 inst = ir_to_mesa_emit_op1(ir, OPCODE_SCS, tmp_dst, src0);
+	 inst->dst_reg.writemask = scs_mask;
+
+	 /* Move the result of the SCS instruction to the desired location in
+	  * the destination.
+	  */
+	 tmp.swizzle = MAKE_SWIZZLE4(component, component,
+				     component, component);
+	 inst = ir_to_mesa_emit_op1(ir, OPCODE_SCS, dst, tmp);
+	 inst->dst_reg.writemask = this_mask;
+      } else {
+	 /* Emit the SCS instruction to write directly to the destination.
+	  */
+	 ir_to_mesa_instruction *inst =
+	    ir_to_mesa_emit_op1(ir, OPCODE_SCS, dst, src0);
+	 inst->dst_reg.writemask = scs_mask;
+      }
+
+      done_mask |= this_mask;
+   }
+}
+
 struct ir_to_mesa_src_reg
 ir_to_mesa_visitor::src_reg_for_float(float val)
 {
@@ -942,6 +1046,12 @@ ir_to_mesa_visitor::visit(ir_expression *ir)
    case ir_unop_cos:
       ir_to_mesa_emit_scalar_op1(ir, OPCODE_COS, result_dst, op[0]);
       break;
+   case ir_unop_sin_reduced:
+      emit_scs(ir, OPCODE_SIN, result_dst, op[0]);
+      break;
+   case ir_unop_cos_reduced:
+      emit_scs(ir, OPCODE_COS, result_dst, op[0]);
+      break;
 
    case ir_unop_dFdx:
       ir_to_mesa_emit_op1(ir, OPCODE_DDX, result_dst, op[0]);