diff options
author | Rhys Perry <[email protected]> | 2019-10-08 14:47:00 +0200 |
---|---|---|
committer | Timur Kristóf <[email protected]> | 2019-10-10 09:57:53 +0200 |
commit | d63c175897e31e2a78eb346362a7c90a2fec5f13 (patch) | |
tree | a32db5d4f82beffd2e6c05e4f55cac57744be27e /src | |
parent | 83993f535eb90874ca2256ddbd35bce4e407c13a (diff) |
aco: pad code with s_code_end on GFX10
Signed-off-by: Rhys Perry <[email protected]>
Reviewed-By: Timur Kristóf <[email protected]>
Reviewed-by: Daniel Schürmann <[email protected]>
Diffstat (limited to 'src')
-rw-r--r-- | src/amd/compiler/aco_assembler.cpp | 15 |
1 files changed, 13 insertions, 2 deletions
diff --git a/src/amd/compiler/aco_assembler.cpp b/src/amd/compiler/aco_assembler.cpp index 5a82d44d74b..76361241e4b 100644 --- a/src/amd/compiler/aco_assembler.cpp +++ b/src/amd/compiler/aco_assembler.cpp @@ -4,6 +4,7 @@ #include "aco_ir.h" #include "common/sid.h" #include "ac_shader_util.h" +#include "util/u_math.h" namespace aco { @@ -668,16 +669,26 @@ unsigned emit_program(Program* program, } fix_branches(ctx, code); + + unsigned exec_size = code.size() * sizeof(uint32_t); + + if (program->chip_class >= GFX10) { + /* Pad output with s_code_end so instruction prefetching doesn't cause + * page faults */ + unsigned final_size = align(code.size() + 3 * 16, 16); + while (code.size() < final_size) + code.push_back(0xbf9f0000u); + } + fix_constaddrs(ctx, code); - unsigned constant_data_offset = code.size() * sizeof(uint32_t); while (program->constant_data.size() % 4u) program->constant_data.push_back(0); /* Copy constant data */ code.insert(code.end(), (uint32_t*)program->constant_data.data(), (uint32_t*)(program->constant_data.data() + program->constant_data.size())); - return constant_data_offset; + return exec_size; } } |