glsl: Implement [iu]mulExtended() built-ins for ARB_gpu_shader5.

These built-ins have two "out" parameters, which makes implementing them
efficiently with our current compiler infrastructure difficult. Instead,
implement them in terms of the existing ir_binop_mul IR (to return the
low 32-bits) and a new ir_binop_mul64 which returns the high 32-bits.

v2: Rename mul64 -> imul_high as suggested by Ken.
Reviewed-by: Kenneth Graunke <[email protected]>

8 files changed, 49 insertions, 1 deletions

diff --git a/src/glsl/builtin_functions.cpp b/src/glsl/builtin_functions.cpp
index 5b1b9c3b7c1..d40888d3819 100644
--- a/src/glsl/builtin_functions.cpp
+++ b/src/glsl/builtin_functions.cpp
@@ -533,6 +533,7 @@ private:
    B2(frexp)
    B1(uaddCarry)
    B1(usubBorrow)
+   B1(mulExtended)
 #undef B0
 #undef B1
 #undef B2
@@ -1961,6 +1962,18 @@ builtin_builder::create_builtins()
                 _usubBorrow(glsl_type::uvec3_type),
                 _usubBorrow(glsl_type::uvec4_type),
                 NULL);
+   add_function("imulExtended",
+                _mulExtended(glsl_type::int_type),
+                _mulExtended(glsl_type::ivec2_type),
+                _mulExtended(glsl_type::ivec3_type),
+                _mulExtended(glsl_type::ivec4_type),
+                NULL);
+   add_function("umulExtended",
+                _mulExtended(glsl_type::uint_type),
+                _mulExtended(glsl_type::uvec2_type),
+                _mulExtended(glsl_type::uvec3_type),
+                _mulExtended(glsl_type::uvec4_type),
+                NULL);
 #undef F
 #undef FI
 #undef FIU
@@ -3762,6 +3775,24 @@ builtin_builder::_usubBorrow(const glsl_type *type)
 
    return sig;
 }
+
+/**
+ * For both imulExtended() and umulExtended() built-ins.
+ */
+ir_function_signature *
+builtin_builder::_mulExtended(const glsl_type *type)
+{
+   ir_variable *x = in_var(type, "x");
+   ir_variable *y = in_var(type, "y");
+   ir_variable *msb = out_var(type, "msb");
+   ir_variable *lsb = out_var(type, "lsb");
+   MAKE_SIG(glsl_type::void_type, gpu_shader5, 4, x, y, msb, lsb);
+
+   body.emit(assign(msb, imul_high(x, y)));
+   body.emit(assign(lsb, mul(x, y)));
+
+   return sig;
+}
 /** @} */
 
 /******************************************************************************/
diff --git a/src/glsl/ir.cpp b/src/glsl/ir.cpp
index 149ddbd5342..ead7456dfcc 100644
--- a/src/glsl/ir.cpp
+++ b/src/glsl/ir.cpp
@@ -398,6 +398,7 @@ ir_expression::ir_expression(int op, ir_rvalue *op0, ir_rvalue *op1)
       this->type = glsl_type::uint_type;
       break;
 
+   case ir_binop_imul_high:
    case ir_binop_carry:
    case ir_binop_borrow:
    case ir_binop_lshift:
@@ -529,6 +530,7 @@ static const char *const operator_strs[] = {
    "+",
    "-",
    "*",
+   "imul_high",
    "/",
    "carry",
    "borrow",
diff --git a/src/glsl/ir.h b/src/glsl/ir.h
index 756ce9c7cd3..9fd5f5a997a 100644
--- a/src/glsl/ir.h
+++ b/src/glsl/ir.h
@@ -1091,7 +1091,8 @@ enum ir_expression_operation {
 
    ir_binop_add,
    ir_binop_sub,
-   ir_binop_mul,
+   ir_binop_mul,       /**< Floating-point or low 32-bit integer multiply. */
+   ir_binop_imul_high, /**< Calculates the high 32-bits of a 64-bit multiply. */
    ir_binop_div,
 
    /**
diff --git a/src/glsl/ir_builder.cpp b/src/glsl/ir_builder.cpp
index b6ce889834c..6c49734bef8 100644
--- a/src/glsl/ir_builder.cpp
+++ b/src/glsl/ir_builder.cpp
@@ -216,6 +216,11 @@ ir_expression *mul(operand a, operand b)
    return expr(ir_binop_mul, a, b);
 }
 
+ir_expression *imul_high(operand a, operand b)
+{
+   return expr(ir_binop_imul_high, a, b);
+}
+
 ir_expression *div(operand a, operand b)
 {
    return expr(ir_binop_div, a, b);
diff --git a/src/glsl/ir_builder.h b/src/glsl/ir_builder.h
index 1345788ab13..1f0778870e3 100644
--- a/src/glsl/ir_builder.h
+++ b/src/glsl/ir_builder.h
@@ -133,6 +133,7 @@ ir_expression *expr(ir_expression_operation op, operand a, operand b, operand c)
 ir_expression *add(operand a, operand b);
 ir_expression *sub(operand a, operand b);
 ir_expression *mul(operand a, operand b);
+ir_expression *imul_high(operand a, operand b);
 ir_expression *div(operand a, operand b);
 ir_expression *carry(operand a, operand b);
 ir_expression *borrow(operand a, operand b);
diff --git a/src/glsl/ir_validate.cpp b/src/glsl/ir_validate.cpp
index 38db8c20677..d6abc8eb5a8 100644
--- a/src/glsl/ir_validate.cpp
+++ b/src/glsl/ir_validate.cpp
@@ -429,6 +429,12 @@ ir_validate::visit_leave(ir_expression *ir)
       }
       break;
 
+   case ir_binop_imul_high:
+      assert(ir->type == ir->operands[0]->type);
+      assert(ir->type == ir->operands[1]->type);
+      assert(ir->type->is_integer());
+      break;
+
    case ir_binop_carry:
    case ir_binop_borrow:
       assert(ir->type == ir->operands[0]->type);
diff --git a/src/mesa/program/ir_to_mesa.cpp b/src/mesa/program/ir_to_mesa.cpp
index 2f87cfd65c4..cfad91bc423 100644
--- a/src/mesa/program/ir_to_mesa.cpp
+++ b/src/mesa/program/ir_to_mesa.cpp
@@ -1499,6 +1499,7 @@ ir_to_mesa_visitor::visit(ir_expression *ir)
    case ir_triop_csel:
    case ir_binop_carry:
    case ir_binop_borrow:
+   case ir_binop_imul_high:
       assert(!"not supported");
       break;
 
diff --git a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
index f33bea6eb02..8d06ac6f7a7 100644
--- a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
+++ b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
@@ -1980,6 +1980,7 @@ glsl_to_tgsi_visitor::visit(ir_expression *ir)
    case ir_binop_ldexp:
    case ir_binop_carry:
    case ir_binop_borrow:
+   case ir_binop_imul_high:
       /* This operation is not supported, or should have already been handled.
        */
       assert(!"Invalid ir opcode in glsl_to_tgsi_visitor::visit()");