i965: Add an implemnetation of nir_op_fquantize2f16

Reviewed-by: Matt Turner <mattst88@gmail.com>

diff --git a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
index 4de559941ceff38b49a36d8c82cd667d8e67a368..7839428c52e644dc205695c479ddc79f2d9472ec 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
@@ -993,6 +993,34 @@ fs_visitor::nir_emit_alu(const fs_builder &bld, nir_alu_instr *instr)
       inst->saturate = instr->dest.saturate;
       break;
 
+   case nir_op_fquantize2f16: {
+      fs_reg tmp16 = bld.vgrf(BRW_REGISTER_TYPE_D);
+      fs_reg tmp32 = bld.vgrf(BRW_REGISTER_TYPE_F);
+      fs_reg zero = bld.vgrf(BRW_REGISTER_TYPE_F);
+
+      /* The destination stride must be at least as big as the source stride. */
+      tmp16.type = BRW_REGISTER_TYPE_W;
+      tmp16.stride = 2;
+
+      /* Check for denormal */
+      fs_reg abs_src0 = op[0];
+      abs_src0.abs = true;
+      bld.CMP(bld.null_reg_f(), abs_src0, brw_imm_f(ldexpf(1.0, -14)),
+              BRW_CONDITIONAL_L);
+      /* Get the appropriately signed zero */
+      bld.AND(retype(zero, BRW_REGISTER_TYPE_UD),
+              retype(op[0], BRW_REGISTER_TYPE_UD),
+              brw_imm_ud(0x80000000));
+      /* Do the actual F32 -> F16 -> F32 conversion */
+      bld.emit(BRW_OPCODE_F32TO16, tmp16, op[0]);
+      bld.emit(BRW_OPCODE_F16TO32, tmp32, tmp16);
+      /* Select that or zero based on normal status */
+      inst = bld.SEL(result, zero, tmp32);
+      inst->predicate = BRW_PREDICATE_NORMAL;
+      inst->saturate = instr->dest.saturate;
+      break;
+   }
+
    case nir_op_fmin:
    case nir_op_imin:
    case nir_op_umin:

diff --git a/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp b/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp
index eef3940b643c7206f08ecd2e7b93c6f3a6d5235a..ee6929b16a2ed16c4f7f0cfdca9321f4413fbc8a 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp
+++ b/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp
@@ -1188,6 +1188,31 @@ vec4_visitor::nir_emit_alu(nir_alu_instr *instr)
       inst->saturate = instr->dest.saturate;
       break;
 
+   case nir_op_fquantize2f16: {
+      /* See also vec4_visitor::emit_pack_half_2x16() */
+      src_reg tmp16 = src_reg(this, glsl_type::uvec4_type);
+      src_reg tmp32 = src_reg(this, glsl_type::vec4_type);
+      src_reg zero = src_reg(this, glsl_type::vec4_type);
+
+      /* Check for denormal */
+      src_reg abs_src0 = op[0];
+      abs_src0.abs = true;
+      emit(CMP(dst_null_f(), abs_src0, brw_imm_f(ldexpf(1.0, -14)),
+               BRW_CONDITIONAL_L));
+      /* Get the appropriately signed zero */
+      emit(AND(retype(dst_reg(zero), BRW_REGISTER_TYPE_UD),
+               retype(op[0], BRW_REGISTER_TYPE_UD),
+               brw_imm_ud(0x80000000)));
+      /* Do the actual F32 -> F16 -> F32 conversion */
+      emit(F32TO16(dst_reg(tmp16), op[0]));
+      emit(F16TO32(dst_reg(tmp32), tmp16));
+      /* Select that or zero based on normal status */
+      inst = emit(BRW_OPCODE_SEL, dst, zero, tmp32);
+      inst->predicate = BRW_PREDICATE_NORMAL;
+      inst->saturate = instr->dest.saturate;
+      break;
+   }
+
    case nir_op_fmin:
    case nir_op_imin:
    case nir_op_umin: