aco: implement 16-bit nir_op_f2i64/nir_op_f2u64

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4509>

diff --git a/src/amd/compiler/aco_instruction_selection.cpp b/src/amd/compiler/aco_instruction_selection.cpp
index 1981b8ec9318a8ac36d10da17ab4d5421a0e619d..f713a7610083ba4330f6eb826e9835c4de3b312b 100644 (file)
--- a/src/amd/compiler/aco_instruction_selection.cpp
+++ b/src/amd/compiler/aco_instruction_selection.cpp
@@ -2351,7 +2351,10 @@ void visit_alu_instr(isel_context *ctx, nir_alu_instr *instr)
    }
    case nir_op_f2i64: {
       Temp src = get_alu_src(ctx, instr->src[0]);
-      if (instr->src[0].src.ssa->bit_size == 32 && dst.type() == RegType::vgpr) {
+      if (instr->src[0].src.ssa->bit_size == 16)
+         src = bld.vop1(aco_opcode::v_cvt_f32_f16, bld.def(v1), src);
+
+      if (instr->src[0].src.ssa->bit_size <= 32 && dst.type() == RegType::vgpr) {
          Temp exponent = bld.vop1(aco_opcode::v_frexp_exp_i32_f32, bld.def(v1), src);
          exponent = bld.vop3(aco_opcode::v_med3_i32, bld.def(v1), Operand(0x0u), exponent, Operand(64u));
          Temp mantissa = bld.vop2(aco_opcode::v_and_b32, bld.def(v1), Operand(0x7fffffu), src);
@@ -2377,7 +2380,7 @@ void visit_alu_instr(isel_context *ctx, nir_alu_instr *instr)
          Temp new_upper = bld.vsub32(bld.def(v1), upper, sign, false, borrow);
          bld.pseudo(aco_opcode::p_create_vector, Definition(dst), new_lower, new_upper);
 
-      } else if (instr->src[0].src.ssa->bit_size == 32 && dst.type() == RegType::sgpr) {
+      } else if (instr->src[0].src.ssa->bit_size <= 32 && dst.type() == RegType::sgpr) {
          if (src.type() == RegType::vgpr)
             src = bld.as_uniform(src);
          Temp exponent = bld.sop2(aco_opcode::s_bfe_u32, bld.def(s1), bld.def(s1, scc), src, Operand(0x80017u));
@@ -2427,7 +2430,10 @@ void visit_alu_instr(isel_context *ctx, nir_alu_instr *instr)
    }
    case nir_op_f2u64: {
       Temp src = get_alu_src(ctx, instr->src[0]);
-      if (instr->src[0].src.ssa->bit_size == 32 && dst.type() == RegType::vgpr) {
+      if (instr->src[0].src.ssa->bit_size == 16)
+         src = bld.vop1(aco_opcode::v_cvt_f32_f16, bld.def(v1), src);
+
+      if (instr->src[0].src.ssa->bit_size <= 32 && dst.type() == RegType::vgpr) {
          Temp exponent = bld.vop1(aco_opcode::v_frexp_exp_i32_f32, bld.def(v1), src);
          Temp exponent_in_range = bld.vopc(aco_opcode::v_cmp_ge_i32, bld.hint_vcc(bld.def(bld.lm)), Operand(64u), exponent);
          exponent = bld.vop2(aco_opcode::v_max_i32, bld.def(v1), Operand(0x0u), exponent);
@@ -2450,7 +2456,7 @@ void visit_alu_instr(isel_context *ctx, nir_alu_instr *instr)
          upper = bld.vop2(aco_opcode::v_cndmask_b32, bld.def(v1), Operand(0xffffffffu), upper, exponent_in_range);
          bld.pseudo(aco_opcode::p_create_vector, Definition(dst), lower, upper);
 
-      } else if (instr->src[0].src.ssa->bit_size == 32 && dst.type() == RegType::sgpr) {
+      } else if (instr->src[0].src.ssa->bit_size <= 32 && dst.type() == RegType::sgpr) {
          if (src.type() == RegType::vgpr)
             src = bld.as_uniform(src);
          Temp exponent = bld.sop2(aco_opcode::s_bfe_u32, bld.def(s1), bld.def(s1, scc), src, Operand(0x80017u));