Mesa (master): aco: flush denormals before fp16 fabs/fneg if needed

[GitLab Mirror]

Module: Mesa
Branch: master
Commit: d14f4faa1378c559af71136cb084e964846e69c5
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=d14f4faa1378c559af71136cb084e964846e69c5

Author: Rhys Perry <pendingchaos02 at gmail.com>
Date:   Thu Jul  2 13:34:57 2020 +0100

aco: flush denormals before fp16 fabs/fneg if needed

Signed-off-by: Rhys Perry <pendingchaos02 at gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5773>

---

 src/amd/compiler/aco_instruction_selection.cpp | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/src/amd/compiler/aco_instruction_selection.cpp b/src/amd/compiler/aco_instruction_selection.cpp
index 138cf7de909..5421bab2c71 100644
--- a/src/amd/compiler/aco_instruction_selection.cpp
+++ b/src/amd/compiler/aco_instruction_selection.cpp
@@ -1953,6 +1953,8 @@ void visit_alu_instr(isel_context *ctx, nir_alu_instr *instr)
    case nir_op_fneg: {
       Temp src = get_alu_src(ctx, instr->src[0]);
       if (dst.regClass() == v2b) {
+         if (ctx->block->fp_mode.must_flush_denorms16_64)
+            src = bld.vop2(aco_opcode::v_mul_f16, bld.def(v2b), Operand((uint16_t)0x3C00), as_vgpr(ctx, src));
          bld.vop2(aco_opcode::v_xor_b32, Definition(dst), Operand(0x8000u), as_vgpr(ctx, src));
       } else if (dst.regClass() == v1) {
          if (ctx->block->fp_mode.must_flush_denorms32)
@@ -1975,6 +1977,8 @@ void visit_alu_instr(isel_context *ctx, nir_alu_instr *instr)
    case nir_op_fabs: {
       Temp src = get_alu_src(ctx, instr->src[0]);
       if (dst.regClass() == v2b) {
+         if (ctx->block->fp_mode.must_flush_denorms16_64)
+            src = bld.vop2(aco_opcode::v_mul_f16, bld.def(v2b), Operand((uint16_t)0x3C00), as_vgpr(ctx, src));
          bld.vop2(aco_opcode::v_and_b32, Definition(dst), Operand(0x7FFFu), as_vgpr(ctx, src));
       } else if (dst.regClass() == v1) {
          if (ctx->block->fp_mode.must_flush_denorms32)