i965/fs: Fix 32-bit integer multiplication.

The MUL opcode does a 16bit * 32bit multiply, and we need to do the
MACH to get the top 16bit * 32bit added in.

Fixes fs-op-mult-int-*, fs-op-mult-ivec*

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This commit is contained in:
Eric Anholt
2011-08-15 22:36:18 -07:00
parent 34980cd153
commit 3f78f71973
2 changed files with 22 additions and 1 deletions

View File

@@ -630,6 +630,11 @@ fs_visitor::generate_code()
case BRW_OPCODE_MUL:
brw_MUL(p, dst, src[0], src[1]);
break;
case BRW_OPCODE_MACH:
brw_set_acc_write_control(p, 1);
brw_MACH(p, dst, src[0], src[1]);
brw_set_acc_write_control(p, 0);
break;
case BRW_OPCODE_FRC:
brw_FRC(p, dst, src[0]);

View File

@@ -287,7 +287,23 @@ fs_visitor::visit(ir_expression *ir)
break;
case ir_binop_mul:
emit(BRW_OPCODE_MUL, this->result, op[0], op[1]);
if (ir->type->is_integer()) {
/* For integer multiplication, the MUL uses the low 16 bits
* of one of the operands (src0 on gen6, src1 on gen7). The
* MACH accumulates in the contribution of the upper 16 bits
* of that operand.
*
* FINISHME: Emit just the MUL if we know an operand is small
* enough.
*/
struct brw_reg acc = retype(brw_acc_reg(), BRW_REGISTER_TYPE_D);
emit(BRW_OPCODE_MUL, acc, op[0], op[1]);
emit(BRW_OPCODE_MACH, reg_null_d, op[0], op[1]);
emit(BRW_OPCODE_MOV, this->result, fs_reg(acc));
} else {
emit(BRW_OPCODE_MUL, this->result, op[0], op[1]);
}
break;
case ir_binop_div:
assert(!"not reached: should be handled by ir_div_to_mul_rcp");