aco: optimize 16-bit and 64-bit float comparisons

[mesa.git] / src / amd / compiler / aco_assembler.cpp

diff --git a/src/amd/compiler/aco_assembler.cpp b/src/amd/compiler/aco_assembler.cpp
index 8876b5bc8f8047ed3c2f64f841930c430e399c3c..3bdea63201d0c893bcd737ce80da0e9962b15588 100644 (file)
--- a/src/amd/compiler/aco_assembler.cpp
+++ b/src/amd/compiler/aco_assembler.cpp
@@ -42,8 +42,6 @@ static uint32_t get_sdwa_sel(unsigned sel, PhysReg reg)
 
 void emit_instruction(asm_context& ctx, std::vector<uint32_t>& out, Instruction* instr)
 {
-   uint32_t instr_offset = out.size() * 4u;
-
    /* lower remaining pseudo-instructions */
    if (instr->opcode == aco_opcode::p_constaddr) {
       unsigned dest = instr->definitions[0].physReg();
@@ -68,7 +66,7 @@ void emit_instruction(asm_context& ctx, std::vector<uint32_t>& out, Instruction*
       encoding |= 255 << 8;
       out.push_back(encoding);
       ctx.constaddrs.push_back(out.size());
-      out.push_back(-(instr_offset + 4) + offset);
+      out.push_back(offset);
 
       /* s_addc_u32 dest[1], dest[1], 0 */
       encoding = (0b10 << 30);
@@ -797,7 +795,7 @@ void fix_branches(asm_context& ctx, std::vector<uint32_t>& out)
 void fix_constaddrs(asm_context& ctx, std::vector<uint32_t>& out)
 {
    for (unsigned addr : ctx.constaddrs)
-      out[addr] += out.size() * 4u;
+      out[addr] += (out.size() - addr + 1u) * 4u;
 }
 
 unsigned emit_program(Program* program,