int64_t kernel_code_prefetch_byte_offset;
uint64_t kernel_code_prefetch_byte_size;
- /// Number of bytes of scratch backing memory required for full
- /// occupancy of target chip. This takes into account the number of
- /// bytes of scratch per work-item, the wavefront size, the maximum
- /// number of wavefronts per CU, and the number of CUs. This is an
- /// upper limit on scratch. If the grid being dispatched is small it
- /// may only need less than this. If the kernel uses no scratch, or
- /// the Finalizer has not computed this value, it must be 0.
- uint64_t max_scratch_backing_memory_byte_size;
+ /// Reserved. Must be 0.
+ uint64_t reserved0;
/// Shader program settings for CS. Contains COMPUTE_PGM_RSRC1 and
/// COMPUTE_PGM_RSRC2 registers.
bool AMDGPUAsmParser::ParseAMDKernelCodeTValue(StringRef ID,
amd_kernel_code_t &Header) {
+ // max_scratch_backing_memory_byte_size is deprecated. Ignore it while parsing
+ // assembly for backwards compatibility.
+ if (ID == "max_scratch_backing_memory_byte_size") {
+ Parser.eatToEndOfStatement();
+ return false;
+ }
+
SmallString<40> ErrStr;
raw_svector_ostream Err(ErrStr);
if (!parseAmdKernelCodeField(ID, getParser(), Header, Err)) {
memset(&Header, 0, sizeof(Header));
Header.amd_kernel_code_version_major = 1;
- Header.amd_kernel_code_version_minor = 1;
+ Header.amd_kernel_code_version_minor = 2;
Header.amd_machine_kind = 1; // AMD_MACHINE_KIND_AMDGPU
Header.amd_machine_version_major = ISA.Major;
Header.amd_machine_version_minor = ISA.Minor;
FIELD(kernel_code_entry_byte_offset),
FIELD(kernel_code_prefetch_byte_size),
-FIELD(max_scratch_backing_memory_byte_size),
COMPPGM1(granulated_workitem_vgpr_count, compute_pgm_rsrc1_vgprs, VGPRS),
COMPPGM1(granulated_wavefront_sgpr_count, compute_pgm_rsrc1_sgprs, SGPRS),
// ASM-LABEL: {{^}}amd_kernel_code_t_minimal:
// ASM: .amd_kernel_code_t
// ASM: amd_code_version_major = 7
-// ASM: amd_code_version_minor = 1
+// ASM: amd_code_version_minor = 2
// ASM: amd_machine_kind = 1
// ASM: amd_machine_version_major = 7
// ASM: amd_machine_version_minor = 0
// ASM: amd_machine_version_stepping = 0
// ASM: kernel_code_entry_byte_offset = 256
// ASM: kernel_code_prefetch_byte_size = 0
-// ASM: max_scratch_backing_memory_byte_size = 0
// ASM: granulated_workitem_vgpr_count = 1
// ASM: granulated_wavefront_sgpr_count = 1
// ASM: priority = 0
// ASM: amd_machine_version_stepping = 5
// ASM: kernel_code_entry_byte_offset = 512
// ASM: kernel_code_prefetch_byte_size = 1
-// ASM: max_scratch_backing_memory_byte_size = 1
// ASM: granulated_workitem_vgpr_count = 1
// ASM: granulated_wavefront_sgpr_count = 1
// ASM: priority = 1
// ASM-LABEL: {{^}}amd_kernel_code_t_minimal:
// ASM: .amd_kernel_code_t
// ASM: amd_code_version_major = 1
-// ASM: amd_code_version_minor = 1
+// ASM: amd_code_version_minor = 2
// ASM: amd_machine_kind = 1
// ASM: amd_machine_version_major = 7
// ASM: amd_machine_version_minor = 0
// ASM: amd_machine_version_stepping = 0
// ASM: kernel_code_entry_byte_offset = 256
// ASM: kernel_code_prefetch_byte_size = 0
-// ASM: max_scratch_backing_memory_byte_size = 0
// ASM: granulated_workitem_vgpr_count = 1
// ASM: granulated_wavefront_sgpr_count = 1
// ASM: priority = 0