Skip to content

spirv-opt: Fix assertion with fp8 #6677

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

Open
GinShio wants to merge 1 commit into
base: main
Choose a base branch
from
Open

spirv-opt: Fix assertion with fp8 #6677

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
2 changes: 1 addition & 1 deletion source/opt/folding_rules.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -153,7 +153,7 @@ std::vector<uint32_t> GetWordsFromScalarFloatConstant(
const analysis::FloatConstant* c) {
assert(c != nullptr);
uint32_t width = c->type()->AsFloat()->width();
assert(width == 16 || width == 32 || width == 64);
assert(width == 8 || width == 16 || width == 32 || width == 64);
if (width == 64) {
utils::FloatProxy<double> result(c->GetDouble());
return result.GetWords();
Expand Down
45 changes: 44 additions & 1 deletion test/opt/fold_test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -226,6 +226,7 @@ OpCapability Int16
OpCapability Int64
OpCapability CooperativeMatrixKHR
OpExtension "SPV_KHR_cooperative_matrix"
OpExtension "SPV_EXT_float8"
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint Fragment %main "main"
Expand All @@ -238,6 +239,8 @@ OpName %main "main"
%float = OpTypeFloat 32
%double = OpTypeFloat 64
%half = OpTypeFloat 16
%float8e4m3 = OpTypeFloat 8 Float8E4M3EXT
%float8e5m2 = OpTypeFloat 8 Float8E5M2EXT
%101 = OpConstantTrue %bool ; Need a def with an numerical id to define id maps.
%true = OpConstantTrue %bool
%false = OpConstantFalse %bool
Expand Down Expand Up @@ -444,6 +447,10 @@ OpName %main "main"
%half_null = OpConstantNull %half
%half_0_1 = OpConstantComposite %v2half %108 %half_1
%v4half_0_1_0_0 = OpConstantComposite %v4half %108 %half_1 %108 %108
%float8e4m3_1 = OpConstant %float8e4m3 1.0
%float8e4m3_0x1_7p_0 = OpConstant %float8e4m3 0x1.7p+0 ; float(1.4375) -> float8e4m3(1.5) (RTNE)
%float8e5m2_1 = OpConstant %float8e5m2 1.0
%float8e5m2_0x1_7p_0 = OpConstant %float8e5m2 0x1.7p+0 ; float(1.4375) -> float8e5m2(1.5) (RTNE)
%106 = OpConstantComposite %v4float %float_0 %float_0 %float_0 %float_0
%v4float_0_0_0_0 = OpConstantComposite %v4float %float_0 %float_0 %float_0 %float_0
%v4float_0_0_0_1 = OpConstantComposite %v4float %float_0 %float_0 %float_0 %float_1
Expand Down Expand Up @@ -1346,7 +1353,43 @@ INSTANTIATE_TEST_SUITE_P(TestCase, IntegerInstructionFoldingTest,
"%2 = OpBitReverse %uint %111\n" +
"OpReturn\n" +
"OpFunctionEnd",
2, 0)
2, 0),
// TODO: hex_float.h contains some errors when converting float32 to
// smaller floating point types, which causes incorrect results. Please see
// https://github.com/KhronosGroup/glslang/issues/4241 and
// https://godbolt.org/z/684sEjzGY for details.
// Test case 98: Bit-cast float8e4m3 1 to float8e4m3
InstructionFoldingCase<uint32_t>(
Header() + "%main = OpFunction %void None %void_func\n" +
"%main_lab = OpLabel\n" +
"%2 = OpBitcast %float8e4m3 %float8e4m3_1\n" +
"OpReturn\n" +
"OpFunctionEnd",
2, 0x38),
// Test case 99: Bit-cast float8e4m3 1.4375 to ubyte
InstructionFoldingCase<uint32_t>(
Header() + "%main = OpFunction %void None %void_func\n" +
"%main_lab = OpLabel\n" +
"%2 = OpBitcast %ubyte %float8e4m3_0x1_7p_0\n" +
"OpReturn\n" +
"OpFunctionEnd",
2, 0x3b),
// Test case 100: Bit-cast float8e5m2 1 to float8e5m2
InstructionFoldingCase<uint32_t>(
Header() + "%main = OpFunction %void None %void_func\n" +
"%main_lab = OpLabel\n" +
"%2 = OpBitcast %float8e5m2 %float8e5m2_1\n" +
"OpReturn\n" +
"OpFunctionEnd",
2, 0x3c),
// Test case 101: Bit-cast float8e5m2 1.4375 to ubyte
InstructionFoldingCase<uint32_t>(
Header() + "%main = OpFunction %void None %void_func\n" +
"%main_lab = OpLabel\n" +
"%2 = OpBitcast %ubyte %float8e5m2_0x1_7p_0\n" +
"OpReturn\n" +
"OpFunctionEnd",
2, 0x3d)
));
// clang-format on

Expand Down
Loading