a636928bb4
The expansion converts arbitrary-precision FP represented as integer following these algorithm: 1. Extract sign, exponent, and mantissa bit fields via masks and shifts. 2. Classify the input (zero, denormal, normal, Inf, NaN) using the exponent and mantissa fields. 3. Normal path: adjusting the exponent bias and left-shifting the mantissa to fit the wider destination format. 4. Denormal path: normalizing by finding the MSB position of the mantissa (via count-leading-zeros), computing the correct exponent from that position, stripping the implicit leading 1, and shifting the fraction into the destination mantissa field. 5. Assemble the destination IEEE bit pattern (sign | exponent | mantissa) and select among the normal, denormal, and special-value results. Currently only conversions from OCP floats are covered, in LLVM terms these are: Float8E5M2, Float8E4M3FN, Float6E3M2FN, Float6E2M3FN, Float4E2M1FN. OCP spec: https://www.opencompute.org/documents/ocp-microscaling-formats-mx-v1-0-spec-final-pdf AI has assisted in X86 E2E testing.
77 lines
2.8 KiB
LLVM
77 lines
2.8 KiB
LLVM
; RUN: split-file %s %t
|
|
; RUN: not llc < %t/float8e4m3.ll -mtriple=x86_64-unknown-unknown 2>&1 | FileCheck %s --check-prefix=E4M3
|
|
; RUN: not llc < %t/float8e3m4.ll -mtriple=x86_64-unknown-unknown 2>&1 | FileCheck %s --check-prefix=E3M4
|
|
; RUN: not llc < %t/float8e5m2fnuz.ll -mtriple=x86_64-unknown-unknown 2>&1 | FileCheck %s --check-prefix=E5M2FNUZ
|
|
; RUN: not llc < %t/float8e4m3fnuz.ll -mtriple=x86_64-unknown-unknown 2>&1 | FileCheck %s --check-prefix=E4M3FNUZ
|
|
; RUN: not llc < %t/float8e4m3b11fnuz.ll -mtriple=x86_64-unknown-unknown 2>&1 | FileCheck %s --check-prefix=E4M3B11FNUZ
|
|
; RUN: not llc < %t/float8e8m0fnu.ll -mtriple=x86_64-unknown-unknown 2>&1 | FileCheck %s --check-prefix=E8M0FNU
|
|
|
|
; Test that llvm.convert.from.arbitrary.fp emits an error for formats that pass
|
|
; verifier validation but are not yet implemented in SelectionDAGBuilder.
|
|
|
|
;--- float8e4m3.ll
|
|
; E4M3: error: convert_from_arbitrary_fp: not implemented format 'Float8E4M3'
|
|
|
|
declare float @llvm.convert.from.arbitrary.fp.f32.i8(i8, metadata)
|
|
|
|
define float @from_f8e4m3(i8 %v) {
|
|
%r = call float @llvm.convert.from.arbitrary.fp.f32.i8(
|
|
i8 %v, metadata !"Float8E4M3")
|
|
ret float %r
|
|
}
|
|
|
|
;--- float8e3m4.ll
|
|
; E3M4: error: convert_from_arbitrary_fp: not implemented format 'Float8E3M4'
|
|
|
|
declare float @llvm.convert.from.arbitrary.fp.f32.i8(i8, metadata)
|
|
|
|
define float @from_f8e3m4(i8 %v) {
|
|
%r = call float @llvm.convert.from.arbitrary.fp.f32.i8(
|
|
i8 %v, metadata !"Float8E3M4")
|
|
ret float %r
|
|
}
|
|
|
|
;--- float8e5m2fnuz.ll
|
|
; E5M2FNUZ: error: convert_from_arbitrary_fp: not implemented format 'Float8E5M2FNUZ'
|
|
|
|
declare float @llvm.convert.from.arbitrary.fp.f32.i8(i8, metadata)
|
|
|
|
define float @from_f8e5m2fnuz(i8 %v) {
|
|
%r = call float @llvm.convert.from.arbitrary.fp.f32.i8(
|
|
i8 %v, metadata !"Float8E5M2FNUZ")
|
|
ret float %r
|
|
}
|
|
|
|
;--- float8e4m3fnuz.ll
|
|
; E4M3FNUZ: error: convert_from_arbitrary_fp: not implemented format 'Float8E4M3FNUZ'
|
|
|
|
declare float @llvm.convert.from.arbitrary.fp.f32.i8(i8, metadata)
|
|
|
|
define float @from_f8e4m3fnuz(i8 %v) {
|
|
%r = call float @llvm.convert.from.arbitrary.fp.f32.i8(
|
|
i8 %v, metadata !"Float8E4M3FNUZ")
|
|
ret float %r
|
|
}
|
|
|
|
;--- float8e4m3b11fnuz.ll
|
|
; E4M3B11FNUZ: error: convert_from_arbitrary_fp: not implemented format 'Float8E4M3B11FNUZ'
|
|
|
|
declare float @llvm.convert.from.arbitrary.fp.f32.i8(i8, metadata)
|
|
|
|
define float @from_f8e4m3b11fnuz(i8 %v) {
|
|
%r = call float @llvm.convert.from.arbitrary.fp.f32.i8(
|
|
i8 %v, metadata !"Float8E4M3B11FNUZ")
|
|
ret float %r
|
|
}
|
|
|
|
;--- float8e8m0fnu.ll
|
|
; E8M0FNU: error: convert_from_arbitrary_fp: not implemented format 'Float8E8M0FNU'
|
|
|
|
declare float @llvm.convert.from.arbitrary.fp.f32.i8(i8, metadata)
|
|
|
|
define float @from_f8e8m0fnu(i8 %v) {
|
|
%r = call float @llvm.convert.from.arbitrary.fp.f32.i8(
|
|
i8 %v, metadata !"Float8E8M0FNU")
|
|
ret float %r
|
|
}
|