From: Rhys Perry <pendingchaos02@gmail.com>
Date: Thu, 2 Jul 2020 12:33:55 +0000 (+0100)
Subject: aco: use s_round_mode/s_denorm_mode
X-Git-Url: https://git.libre-soc.org/?p=mesa.git;a=commitdiff_plain;h=305cffa22b35ab74980766f94239143935e6e3f6

aco: use s_round_mode/s_denorm_mode

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

diff --git a/src/amd/compiler/aco_ir.h b/src/amd/compiler/aco_ir.h
index a165a1681ad..8adbd567c01 100644
--- a/src/amd/compiler/aco_ir.h
+++ b/src/amd/compiler/aco_ir.h
@@ -142,6 +142,10 @@ struct float_mode {
           unsigned denorm32:2;
           unsigned denorm16_64:2;
       };
+      struct {
+         uint8_t round:4;
+         uint8_t denorm:4;
+      };
       uint8_t val = 0;
    };
    /* if false, optimizations which may remove infs/nan/-0.0 can be done */
diff --git a/src/amd/compiler/aco_lower_to_hw_instr.cpp b/src/amd/compiler/aco_lower_to_hw_instr.cpp
index f052f7ed792..f3903777257 100644
--- a/src/amd/compiler/aco_lower_to_hw_instr.cpp
+++ b/src/amd/compiler/aco_lower_to_hw_instr.cpp
@@ -1637,6 +1637,21 @@ void handle_operands(std::map<PhysReg, copy_operation>& copy_map, lower_context*
    ctx->program->statistics[statistic_copies] += ctx->instructions.size() - num_instructions_before;
 }
 
+void emit_set_mode(Builder& bld, float_mode new_mode, bool set_round, bool set_denorm)
+{
+   if (bld.program->chip_class >= GFX10) {
+      if (set_round)
+         bld.sopp(aco_opcode::s_round_mode, -1, new_mode.round);
+      if (set_denorm)
+         bld.sopp(aco_opcode::s_denorm_mode, -1, new_mode.denorm);
+   } else if (set_round || set_denorm) {
+      /* "((size - 1) << 11) | register" (MODE is encoded as register 1) */
+      Instruction *instr = bld.sopk(aco_opcode::s_setreg_imm32_b32, Operand(new_mode.val), (7 << 11) | 1).instr;
+      /* has to be a literal */
+      instr->operands[0].setFixed(PhysReg{255});
+   }
+}
+
 void lower_to_hw_instr(Program* program)
 {
    Block *discard_block = NULL;
@@ -1648,23 +1663,23 @@ void lower_to_hw_instr(Program* program)
       ctx.program = program;
       Builder bld(program, &ctx.instructions);
 
-      bool set_mode = i == 0 && block->fp_mode.val != program->config->float_mode;
-      for (unsigned pred : block->linear_preds) {
-         if (program->blocks[pred].fp_mode.val != block->fp_mode.val) {
-            set_mode = true;
-            break;
+      float_mode config_mode;
+      config_mode.val = program->config->float_mode;
+
+      bool set_round = i == 0 && block->fp_mode.round != config_mode.round;
+      bool set_denorm = i == 0 && block->fp_mode.denorm != config_mode.denorm;
+      if (block->kind & block_kind_top_level) {
+         for (unsigned pred : block->linear_preds) {
+            if (program->blocks[pred].fp_mode.round != block->fp_mode.round)
+               set_round = true;
+            if (program->blocks[pred].fp_mode.denorm != block->fp_mode.denorm)
+               set_denorm = true;
          }
       }
-      if (set_mode) {
-         /* only allow changing modes at top-level blocks so this doesn't break
-          * the "jump over empty blocks" optimization */
-         assert(block->kind & block_kind_top_level);
-         uint32_t mode = block->fp_mode.val;
-         /* "((size - 1) << 11) | register" (MODE is encoded as register 1) */
-         Instruction *instr = bld.sopk(aco_opcode::s_setreg_imm32_b32, Operand(mode), (7 << 11) | 1).instr;
-         /* has to be a literal */
-         instr->operands[0].setFixed(PhysReg{255});
-      }
+      /* only allow changing modes at top-level blocks so this doesn't break
+       * the "jump over empty blocks" optimization */
+      assert((!set_round && !set_denorm) || (block->kind & block_kind_top_level));
+      emit_set_mode(bld, block->fp_mode, set_round, set_denorm);
 
       for (size_t j = 0; j < block->instructions.size(); j++) {
          aco_ptr<Instruction>& instr = block->instructions[j];