From c5eda3c746b6d6152279d8687cf7885c38a76e93 Mon Sep 17 00:00:00 2001 From: Rhys Perry Date: Mon, 27 Apr 2020 21:16:15 +0100 Subject: [PATCH] aco: improve clamped integer addition disassembly workaround MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit Make it work with 16-bit and GFX10. Signed-off-by: Rhys Perry Reviewed-by: Daniel Schürmann Part-of: --- src/amd/compiler/aco_print_asm.cpp | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-) diff --git a/src/amd/compiler/aco_print_asm.cpp b/src/amd/compiler/aco_print_asm.cpp index e2dbc5bd8b6..1c152e5b73e 100644 --- a/src/amd/compiler/aco_print_asm.cpp +++ b/src/amd/compiler/aco_print_asm.cpp @@ -137,9 +137,14 @@ void print_asm(Program *program, std::vector& binary, size_t new_pos; const int align_width = 60; - if (!l && program->chip_class == GFX9 && ((binary[pos] & 0xffff8000) == 0xd1348000)) { /* not actually an invalid instruction */ - out << std::left << std::setw(align_width) << std::setfill(' ') << "\tv_add_u32_e64 + clamp"; - new_pos = pos + 2; + if (!l && + ((program->chip_class >= GFX9 && (binary[pos] & 0xffff8000) == 0xd1348000) || /* v_add_u32_e64 + clamp */ + (program->chip_class >= GFX10 && (binary[pos] & 0xffff8000) == 0xd7038000) || /* v_add_u16_e64 + clamp */ + (program->chip_class <= GFX9 && (binary[pos] & 0xffff8000) == 0xd1268000)) /* v_add_u16_e64 + clamp */) { + out << std::left << std::setw(align_width) << std::setfill(' ') << "\tinteger addition + clamp"; + bool has_literal = program->chip_class >= GFX10 && + (((binary[pos+1] & 0x1ff) == 0xff) || (((binary[pos+1] >> 9) & 0x1ff) == 0xff)); + new_pos = pos + 2 + has_literal; } else if (program->chip_class == GFX10 && l == 4 && ((binary[pos] & 0xfe0001ff) == 0x020000f9)) { out << std::left << std::setw(align_width) << std::setfill(' ') << "\tv_cndmask_b32 + sdwa"; new_pos = pos + 2; -- 2.30.2