Skip to content

[rocjitsu] Dispatch operand validation by type #10371

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
kuhar merged 1 commit into from
Aug 19, 2026
Merged

[rocjitsu] Dispatch operand validation by type #10371

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 16 additions & 11 deletions emulation/rocjitsu/lib/python/amdisa/codegen/_generator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11410,7 +11410,7 @@ def gen_operand(self) -> None:
if uses_packed_16bit_sources
else ''
)
selector_validation = ''
selector_validation_cases = []
for selector in sorted(
self.isa_spec.opnd_selectors, key=lambda item: item.operand_type
):
Expand Down Expand Up @@ -11438,11 +11438,19 @@ def gen_operand(self) -> None:
f'(encoding_value >= {lo} && encoding_value <= {hi})'
for lo, hi in intervals
)
selector_validation += (
f' if (opr_type == OperandType::{operand_type} && '
f'!({valid_expr}))\n'
f' defer_encoding_error(EncodingError::{error});\n'
selector_validation_cases.append(
f' case OperandType::{operand_type}:\n'
f' if (!({valid_expr}))\n'
f' defer_encoding_error(EncodingError::{error});\n'
f' break;\n'
)
selector_validation = (
' switch (opr_type) {\n'
+ ''.join(selector_validation_cases)
+ ' default:\n'
+ ' break;\n'
+ ' }\n'
)
packed_16bit_ctor_impl = []
if uses_packed_16bit_sources:
packed_16bit_ctor_impl.append(
Expand Down Expand Up @@ -11492,12 +11500,9 @@ def gen_operand(self) -> None:
cgen.Line(
'Operand::Operand(int size_bits, OperandType opr_type, int encoding_value,\n'
' uint16_t literal16_display_value, bool has_literal16_display)\n'
f' : {operand_base_init}(size_bits, opr_type, encoding_value)'
f'{execution_backend_ctor_init},\n'
' literal16_display_value_(literal16_display_value),\n'
' has_literal16_display_(has_literal16_display) {\n'
f'{selector_validation}'
' is_vgpr_ = is_vgpr_operand_type(opr_type);\n'
' : Operand(size_bits, opr_type, encoding_value) {\n'
' literal16_display_value_ = literal16_display_value;\n'
' has_literal16_display_ = has_literal16_display;\n'
'}'
),
cgen.Line(
Expand Down
66 changes: 46 additions & 20 deletions emulation/rocjitsu/lib/python/amdisa/tests/test_generator_profile_gates.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2173,8 +2173,8 @@ def test_generated_operand_validates_scalar_register_selector_intervals(
(amdgpu_generated_root / arch / 'operand.cpp').read_text().split()
)
assert (
'opr_type==OperandType::OPR_SREG&&'
'!((encoding_value>=0&&encoding_value<=123)||'
'caseOperandType::OPR_SREG:'
'if(!((encoding_value>=0&&encoding_value<=123)||'
'(encoding_value>=125&&encoding_value<=125))' in operand
)
assert (
Expand All @@ -2189,12 +2189,12 @@ def test_generated_operand_validates_scalar_register_selector_variants(
gfx1250_operand = ''.join(
(gfx1250_generated_root / 'operand.cpp').read_text().split()
)
assert 'opr_type==OperandType::OPR_SREG_M0&&' in gfx1250_operand
assert 'caseOperandType::OPR_SREG_M0:' in gfx1250_operand

rdna4_operand = ''.join((rdna4_generated_root / 'operand.cpp').read_text().split())
assert (
'opr_type==OperandType::OPR_SREG_LITERAL&&'
'!((encoding_value>=0&&encoding_value<=124)||'
'caseOperandType::OPR_SREG_LITERAL:'
'if(!((encoding_value>=0&&encoding_value<=124)||'
'(encoding_value>=255&&encoding_value<=255))' in rdna4_operand
)

Expand All @@ -2207,13 +2207,13 @@ def test_generated_operand_validates_scalar_selector_families(
)

assert (
'opr_type==OperandType::OPR_SDST&&'
'!((encoding_value>=0&&encoding_value<=124)||'
'caseOperandType::OPR_SDST:'
'if(!((encoding_value>=0&&encoding_value<=124)||'
'(encoding_value>=126&&encoding_value<=127))' in operand
)
assert (
'opr_type==OperandType::OPR_SSRC_NOLIT&&'
'!((encoding_value>=0&&encoding_value<=124)||'
'caseOperandType::OPR_SSRC_NOLIT:'
'if(!((encoding_value>=0&&encoding_value<=124)||'
'(encoding_value>=126&&encoding_value<=208)||'
'(encoding_value>=235&&encoding_value<=248)||'
'(encoding_value>=251&&encoding_value<=253))' in operand
Expand All @@ -2232,7 +2232,7 @@ def test_generated_operand_validates_barrier_id_selectors(
(rdna4_generated_root, 194),
):
operand = ''.join((root / 'operand.cpp').read_text().split())
assert 'opr_type==OperandType::OPR_SSRC_BARRIER_ID&&' in operand
assert 'caseOperandType::OPR_SSRC_BARRIER_ID:' in operand
assert '(encoding_value>=125&&encoding_value<=125)' in operand
assert '(encoding_value>=128&&encoding_value<=159)' in operand
assert f'(encoding_value>=193&&encoding_value<={negative_max})' in operand
Expand Down Expand Up @@ -2261,9 +2261,9 @@ def test_generated_operand_validates_direct_selector_namespaces(
gfx1250_operand = ''.join(
(gfx1250_generated_root / 'operand.cpp').read_text().split()
)
assert 'opr_type==OperandType::OPR_SRC&&' in gfx1250_operand
assert 'caseOperandType::OPR_SRC:' in gfx1250_operand
assert '(encoding_value>=209&&encoding_value<=229)' not in gfx1250_operand
assert 'opr_type==OperandType::OPR_SRC_VGPR_OR_INLINE&&' in gfx1250_operand
assert 'caseOperandType::OPR_SRC_VGPR_OR_INLINE:' in gfx1250_operand
assert '(encoding_value>=128&&encoding_value<=208)' in gfx1250_operand
assert '(encoding_value>=240&&encoding_value<=248)' in gfx1250_operand
assert '(encoding_value>=256&&encoding_value<=511)' in gfx1250_operand
Expand All @@ -2272,8 +2272,8 @@ def test_generated_operand_validates_direct_selector_namespaces(
(amdgpu_generated_root / 'cdna1' / 'operand.cpp').read_text().split()
)
assert (
'opr_type==OperandType::OPR_SMEM_OFFSET&&'
'!((encoding_value>=0&&encoding_value<=124))' in cdna1_operand
'caseOperandType::OPR_SMEM_OFFSET:'
'if(!((encoding_value>=0&&encoding_value<=124))' in cdna1_operand
)


Expand Down Expand Up @@ -2324,7 +2324,7 @@ def test_gfx1250_generated_operand_validates_lane_selectors(

assert 'OPR_SSRC_LANESEL_POS_INT_MAX = 191' in operand_types
assert '#include "util/except.h"' not in operand
assert 'opr_type == OperandType::OPR_SSRC_LANESEL' in operand
assert 'case OperandType::OPR_SSRC_LANESEL:' in operand
assert 'EncodingError::InvalidLaneSelector' in operand


Expand All @@ -2333,7 +2333,7 @@ def test_gfx1250_generated_operand_rejects_invalid_exec_selector(
):
operand = (gfx1250_generated_root / 'operand.cpp').read_text()
assert '#include "util/except.h"' not in operand
assert 'opr_type == OperandType::OPR_EXEC' in operand
assert 'case OperandType::OPR_EXEC:' in operand
assert '(encoding_value >= 126 && encoding_value <= 126)' in operand
assert 'EncodingError::InvalidExecSelector' in operand

Expand Down Expand Up @@ -3346,7 +3346,7 @@ def test_gfx1250_generated_operand_rejects_reserved_scalar_source_selectors(
):
operand = (gfx1250_generated_root / 'operand.cpp').read_text()

assert 'opr_type == OperandType::OPR_SSRC' in operand
assert 'case OperandType::OPR_SSRC:' in operand
assert '(encoding_value >= 209 && encoding_value <= 229)' not in operand
assert 'EncodingError::InvalidScalarSourceSelector' in operand

Expand Down Expand Up @@ -4289,9 +4289,10 @@ def test_generated_operands_validate_vgpr_source_selectors(
amdgpu_generated_root: Path,
):
validation = (
'if (opr_type == OperandType::OPR_SRC_VGPR && '
'!((encoding_value >= 256 && encoding_value <= 511)))\n'
' defer_encoding_error(EncodingError::InvalidVgprSourceSelector);'
'case OperandType::OPR_SRC_VGPR:\n'
' if (!((encoding_value >= 256 && encoding_value <= 511)))\n'
' defer_encoding_error(EncodingError::InvalidVgprSourceSelector);\n'
' break;'
)
for arch in (
'cdna1',
Expand All @@ -4309,6 +4310,31 @@ def test_generated_operands_validate_vgpr_source_selectors(
assert validation in operand


def test_generated_operand_validation_switch_is_shared_by_constructors(
amdgpu_generated_root: Path,
):
literal16_constructor = (
'Operand::Operand(int size_bits, OperandType opr_type, int encoding_value,\n'
' uint16_t literal16_display_value, bool has_literal16_display)\n'
' : Operand(size_bits, opr_type, encoding_value) {'
)
for arch in (
'cdna1',
'cdna2',
'cdna3',
'cdna4',
'cdna5',
'rdna1',
'rdna2',
'rdna3',
'rdna3_5',
'rdna4',
):
operand = (amdgpu_generated_root / arch / 'operand.cpp').read_text()
assert operand.count('switch (opr_type) {') == 1
assert literal16_constructor in operand


def test_cdna4_mfma_f8f6f4_accepts_standalone_and_prefixed_encodings(
amdgpu_generated_root: Path,
):
Expand Down
Loading
Loading