class SVP64ALUTestCase(TestAccumulatorBase):
def case_1_sv_add(self):
- # adds:
- # 1 = 5 + 9 => 0x5555 = 0x4321 + 0x1234
- # 2 = 6 + 10 => 0x3334 = 0x2223 + 0x1111
+ """lst = ['sv.add 1.v, 5.v, 9.v']
+ adds:
+ 1 = 5 + 9 => 0x5555 = 0x4321 + 0x1234
+ 2 = 6 + 10 => 0x3334 = 0x2223 + 0x1111
+ """
isa = SVP64Asm(['sv.add 1.v, 5.v, 9.v'])
lst = list(isa)
print("listing", lst)
initial_svstate=svstate)
def case_2_sv_add_scalar(self):
- # adds:
- # 1 = 5 + 9 => 0x5555 = 0x4321 + 0x1234
+ """lst = ['sv.add 1, 5, 9']
+ adds:
+ 1 = 5 + 9 => 0x5555 = 0x4321 + 0x1234
+ """
isa = SVP64Asm(['sv.add 1, 5, 9'])
lst = list(isa)
print("listing", lst)
self.add_case(Program(lst, bigendian), initial_regs,
initial_svstate=svstate)
- # This case helps checking the encoding of the Extra field
- # It was built so the v3.0b registers are: 3, 2, 1
- # and the Extra field is: 101.110.111
- # The expected SVP64 register numbers are: 13, 10, 7
- # Any mistake in decoding will probably give a different answer
def case_3_sv_check_extra(self):
- # adds:
- # 13 = 10 + 7 => 0x4242 = 0x1230 + 0x3012
+ """lst = ['sv.add 13.v, 10.v, 7.v']
+ adds:
+ 13 = 10 + 7 => 0x4242 = 0x1230 + 0x3012
+
+ This case helps checking the encoding of the Extra field
+ It was built so the v3.0b registers are: 3, 2, 1
+ and the Extra field is: 101.110.111
+ The expected SVP64 register numbers are: 13, 10, 7
+ Any mistake in decoding will probably give a different answer
+ """
isa = SVP64Asm(['sv.add 13.v, 10.v, 7.v'])
lst = list(isa)
print("listing", lst)
initial_svstate=svstate)
def case_4_sv_add_(self):
- # adds when Rc=1: TODO CRs higher up
- # 1 = 5 + 9 => 0 = -1+1 CR0=0b100
- # 2 = 6 + 10 => 0x3334 = 0x2223+0x1111 CR1=0b010
-
+ """lst = ['sv.add. 1.v, 5.v, 9.v']
+ adds when Rc=1: TODO CRs higher up
+ 1 = 5 + 9 => 0 = -1+1 CR0=0b100
+ 2 = 6 + 10 => 0x3334 = 0x2223+0x1111 CR1=0b010
+ """
isa = SVP64Asm(['sv.add. 1.v, 5.v, 9.v'])
lst = list(isa)
print("listing", lst)
initial_svstate=svstate)
def case_5_sv_check_vl_0(self):
- # adds:
- # 1 = 5 + 9 => 0x5555 = 0x4321 + 0x1234
+ """lst = [
+ 'sv.add 13.v, 10.v, 7.v', # skipped, because VL == 0
+ 'add 1, 5, 9'
+ ]
+ adds:
+ 1 = 5 + 9 => 0x5555 = 0x4321 + 0x1234
+ """
isa = SVP64Asm([
'sv.add 13.v, 10.v, 7.v', # skipped, because VL == 0
'add 1, 5, 9'
# checks that SRCSTEP was reset properly after an SV instruction
def case_6_sv_add_multiple(self):
- # adds:
- # 1 = 5 + 9 => 0x5555 = 0x4321 + 0x1234
- # 2 = 6 + 10 => 0x3334 = 0x2223 + 0x1111
- # 3 = 7 + 11 => 0x4242 = 0x3012 + 0x1230
- # 13 = 10 + 7 => 0x2341 = 0x1111 + 0x1230
- # 14 = 11 + 8 => 0x3012 = 0x3012 + 0x0000
- # 15 = 12 + 9 => 0x1234 = 0x0000 + 0x1234
+ """lst = [
+ 'sv.add 1.v, 5.v, 9.v',
+ 'sv.add 13.v, 10.v, 7.v'
+ ]
+ adds:
+ 1 = 5 + 9 => 0x5555 = 0x4321 + 0x1234
+ 2 = 6 + 10 => 0x3334 = 0x2223 + 0x1111
+ 3 = 7 + 11 => 0x4242 = 0x3012 + 0x1230
+ 13 = 10 + 7 => 0x2341 = 0x1111 + 0x1230
+ 14 = 11 + 8 => 0x3012 = 0x3012 + 0x0000
+ 15 = 12 + 9 => 0x1234 = 0x0000 + 0x1234
+ """
isa = SVP64Asm([
'sv.add 1.v, 5.v, 9.v',
'sv.add 13.v, 10.v, 7.v'
initial_svstate=svstate)
def case_7_sv_add_2(self):
- # adds:
- # 1 = 5 + 9 => 0x5555 = 0x4321 + 0x1234
+ """lst = ['sv.add 1, 5.v, 9.v']
+ adds:
+ 1 = 5 + 9 => 0x5555 = 0x4321 + 0x1234
+ """
# r1 is scalar so ENDS EARLY
isa = SVP64Asm(['sv.add 1, 5.v, 9.v'])
lst = list(isa)
initial_svstate=svstate)
def case_8_sv_add_3(self):
- # adds:
- # 1 = 5 + 9 => 0x5555 = 0x4321+0x1234
- # 2 = 5 + 10 => 0x5432 = 0x4321+0x1111
+ """lst = ['sv.add 1.v, 5, 9.v']
+ adds:
+ 1 = 5 + 9 => 0x5555 = 0x4321+0x1234
+ 2 = 5 + 10 => 0x5432 = 0x4321+0x1111
+ """
isa = SVP64Asm(['sv.add 1.v, 5, 9.v'])
lst = list(isa)
print("listing", lst)
initial_svstate=svstate)
def case_9_sv_extsw_intpred(self):
- # extsb, integer twin-pred mask: source is ~r3 (0b01), dest r3 (0b10)
- # works as follows, where any zeros indicate "skip element"
- # - sources are 9 and 10
- # - dests are 5 and 6
- # - source mask says "pick first element from source (5)
- # - dest mask says "pick *second* element from dest (10)
- #
- # therefore the operation that's carried out is:
- # GPR(10) = extsb(GPR(5))
- #
- # this is a type of back-to-back VREDUCE and VEXPAND but it applies
- # to *operations*, not just MVs like in traditional Vector ISAs
- # ascii graphic:
- #
- # reg num 0 1 2 3 4 5 6 7 8 9 10
- # src ~r3=0b01 Y N
- # |
- # +-----+
- # |
- # dest r3=0b10 N Y
-
- # expected results:
- # r5 = 0x0 dest r3 is 0b10: skip
- # r6 = 0xffff_ffff_ffff_ff91 2nd bit of r3 is 1
+ """lst = ['sv.extsb/sm=~r3/dm=r3 5.v, 9.v']
+
+ extsb, integer twin-pred mask: source is ~r3 (0b01), dest r3 (0b10)
+ works as follows, where any zeros indicate "skip element"
+ - sources are 9 and 10
+ - dests are 5 and 6
+ - source mask says "pick first element from source (5)
+ - dest mask says "pick *second* element from dest (10)
+
+ therefore the operation that's carried out is:
+ GPR(10) = extsb(GPR(5))
+
+ this is a type of back-to-back VREDUCE and VEXPAND but it applies
+ to *operations*, not just MVs like in traditional Vector ISAs
+ ascii graphic:
+
+ reg num 0 1 2 3 4 5 6 7 8 9 10
+ predicate src ~r3=0b01 Y N
+ |
+ +-----+
+ |
+ predicate dest r3=0b10 N Y
+
+ expected results:
+ r5 = 0x0 dest r3 is 0b10: skip
+ r6 = 0xffff_ffff_ffff_ff91 2nd bit of r3 is 1
+ """
isa = SVP64Asm(['sv.extsb/sm=~r3/dm=r3 5.v, 9.v'])
lst = list(isa)
print("listing", lst)
initial_svstate=svstate)
def case_10_intpred_vcompress(self):
- # reg num 0 1 2 3 4 5 6 7 8 9 10 11
- # src r3=0b101 Y N Y
- # | |
- # +-------+ |
- # | +-----------+
- # | |
- # dest always Y Y Y
-
- # expected results:
- # r5 = 0xffff_ffff_ffff_ff90 (from r9)
- # r6 = 0xffff_ffff_ffff_ff92 (from r11)
- # r7 = 0x0 (VL loop runs out before we can use it)
+ """lst = ['sv.extsb/sm=r3 5.v, 9.v']
+
+ reg num 0 1 2 3 4 5 6 7 8 9 10 11
+ predicate src r3=0b101 Y N Y
+ | |
+ +-------+ |
+ | +-----------+
+ | |
+ predicate dest always Y Y Y
+
+ expected results:
+ r5 = 0xffff_ffff_ffff_ff90 (from r9)
+ r6 = 0xffff_ffff_ffff_ff92 (from r11)
+ r7 = 0x0 (VL loop runs out before we can use it)
+ """
isa = SVP64Asm(['sv.extsb/sm=r3 5.v, 9.v'])
lst = list(isa)
print("listing", lst)
initial_svstate=svstate)
def case_11_intpred_vexpand(self):
- # reg num 0 1 2 3 4 5 6 7 8 9 10 11
- # src always Y Y Y
- # | |
- # +-------+ |
- # | +------+
- # | |
- # dest r3=0b101 Y N Y
-
- # expected results:
- # r5 = 0xffff_ffff_ffff_ff90 1st bit of r3 is 1
- # r6 = 0x0 skip
- # r7 = 0xffff_ffff_ffff_ff91 3nd bit of r3 is 1
+ """lst = ['sv.extsb/dm=r3 5.v, 9.v']
+
+ reg num 0 1 2 3 4 5 6 7 8 9 10 11
+ predicate src always Y Y Y
+ | |
+ +-------+ |
+ | +------+
+ | |
+ predicate dest r3=0b101 Y N Y
+
+ expected results:
+ r5 = 0xffff_ffff_ffff_ff90 1st bit of r3 is 1
+ r6 = 0x0 skip
+ r7 = 0xffff_ffff_ffff_ff91 3nd bit of r3 is 1
+ """
isa = SVP64Asm(['sv.extsb/dm=r3 5.v, 9.v'])
lst = list(isa)
print("listing", lst)
initial_svstate=svstate)
def case_12_sv_twinpred(self):
- # reg num 0 1 2 3 4 5 6 7 8 9 10 11
- # src r3=0b101 Y N Y
- # |
- # +-----+
- # |
- # dest ~r3=0b010 N Y N
-
- # expected results:
- # r5 = 0x0 dest ~r3 is 0b010: skip
- # r6 = 0xffff_ffff_ffff_ff90 2nd bit of ~r3 is 1
- # r7 = 0x0 dest ~r3 is 0b010: skip
+ """lst = ['sv.extsb/sm=r3/dm=~r3 5.v, 9.v']
+
+ reg num 0 1 2 3 4 5 6 7 8 9 10 11
+ predicate src r3=0b101 Y N Y
+ |
+ +-----+
+ |
+ predicate dest ~r3=0b010 N Y N
+
+ expected results:
+ r5 = 0x0 dest ~r3 is 0b010: skip
+ r6 = 0xffff_ffff_ffff_ff90 2nd bit of ~r3 is 1
+ r7 = 0x0 dest ~r3 is 0b010: skip
+ """
isa = SVP64Asm(['sv.extsb/sm=r3/dm=~r3 5.v, 9.v'])
lst = list(isa)
print("listing", lst)
self.add_case(Program(lst, bigendian), initial_regs,
initial_svstate=svstate)
- # checks integer predication.
def case_13_sv_predicated_add(self):
- # adds:
- # 1 = 5 + 9 => 0x5555 = 0x4321 + 0x1234
- # 2 = 0 (skipped)
- # 3 = 7 + 11 => 0x4242 = 0x3012 + 0x1230
- #
- # 13 = 0 (skipped)
- # 14 = 11 + 8 => 0xB063 = 0x3012 + 0x8051
- # 15 = 0 (skipped)
+ """lst = [
+ 'sv.add/m=r30 1.v, 5.v, 9.v',
+ 'sv.add/m=~r30 13.v, 10.v, 7.v'
+ ]
+
+ checks integer predication using mask-invertmask.
+ real-world usage would be two different operations
+ (a masked-add and an inverted-masked-sub, where the
+ mask was set up as part of a parallel If-Then-Else)
+
+ first add:
+ 1 = 5 + 9 => 0x5555 = 0x4321 + 0x1234
+ 2 = 0 (skipped)
+ 3 = 7 + 11 => 0x4242 = 0x3012 + 0x1230
+
+ second add:
+ 13 = 0 (skipped)
+ 14 = 11 + 8 => 0xB063 = 0x3012 + 0x8051
+ 15 = 0 (skipped)
+ """
isa = SVP64Asm([
'sv.add/m=r30 1.v, 5.v, 9.v',
'sv.add/m=~r30 13.v, 10.v, 7.v'
self.add_case(Program(lst, bigendian), initial_regs,
initial_svstate=svstate)
- # checks an instruction with no effect (all mask bits are zeros)
def case_14_intpred_all_zeros_all_ones(self):
- # adds:
- # 1 = 0 (skipped)
- # 2 = 0 (skipped)
- # 3 = 0 (skipped)
- #
- # 13 = 10 + 7 => 0x2341 = 0x1111 + 0x1230
- # 14 = 11 + 8 => 0xB063 = 0x3012 + 0x8051
- # 15 = 12 + 9 => 0x7736 = 0x6502 + 0x1234
+ """lst = [
+ 'sv.add/m=r30 1.v, 5.v, 9.v',
+ 'sv.add/m=~r30 13.v, 10.v, 7.v'
+ ]
+
+ checks an instruction with no effect (all mask bits are zeros).
+ TODO: check completion time (number of cycles), although honestly
+ it is an implementation-specific optimisation to decide to skip
+ Vector operations with a fully-zero mask.
+
+ first add:
+ 1 = 0 (skipped)
+ 2 = 0 (skipped)
+ 3 = 0 (skipped)
+
+ second add:
+ 13 = 10 + 7 => 0x2341 = 0x1111 + 0x1230
+ 14 = 11 + 8 => 0xB063 = 0x3012 + 0x8051
+ 15 = 12 + 9 => 0x7736 = 0x6502 + 0x1234
+ """
isa = SVP64Asm([
'sv.add/m=r30 1.v, 5.v, 9.v',
'sv.add/m=~r30 13.v, 10.v, 7.v'
self.add_case(Program(lst, bigendian), initial_regs,
initial_svstate=svstate)
- # checks that we are able to resume in the middle of a VL loop,
- # after an interrupt, or after the user has updated src/dst step
- # let's assume the user has prepared src/dst step before running this
- # vector instruction
- # make sure we skip mask bits before the initial step
def case_15_intpred_reentrant(self):
- # reg num 0 1 2 3 4 5 6 7 8 9 10 11 12
- # srcstep=1 v
- # src r3=0b0101 Y N Y N
- # : |
- # + - - + |
- # : +-------+
- # : |
- # dest ~r3=0b1010 N Y N Y
- # dststep=2 ^
- #
- # expected results:
- # r5 = 0x0 # skip
- # r6 = 0x0 # dststep starts at 3, so this gets skipped
- # r7 = 0x0 # skip
- # r8 = 0xffff_ffff_ffff_ff92 # this will be used
-
+ """lst = ['sv.extsb/sm=r3/dm=~r3 5.v, 9.v']
+
+ checks that we are able to resume in the middle of a VL loop,
+ after an interrupt, or after the user has updated src/dst step
+ let's assume the user has prepared src/dst step before running this
+ vector instruction. this is legal but unusual: normally it would
+ be an interrupt return that would have non-zero step values
+
+ note to hardware implementors: inside the hardware,
+ make sure to skip mask bits before the initial step,
+ to save clock cycles. or not. your choice.
+
+ reg num 0 1 2 3 4 5 6 7 8 9 10 11 12
+ srcstep=1 v
+ src r3=0b0101 Y N Y N
+ : |
+ + - - + |
+ : +-------+
+ : |
+ dest ~r3=0b1010 N Y N Y
+ dststep=2 ^
+
+ expected results:
+ r5 = 0x0 # skip
+ r6 = 0x0 # dststep starts at 3, so this gets skipped
+ r7 = 0x0 # skip
+ r8 = 0xffff_ffff_ffff_ff92 # this will be used
+ """
isa = SVP64Asm(['sv.extsb/sm=r3/dm=~r3 5.v, 9.v'])
lst = list(isa)
print("listing", lst)
initial_svstate=svstate)
def case_16_shift_one_by_r3_dest(self):
- # reg num 0 1 2 3 4 5 6 7 8 9 10 11
- # src r30=0b100 N N Y
- # |
- # +-----------+
- # |
- # dest r3=1: 1<<r3=0b010 N Y N
- #
- # expected results:
- # r5 = 0x0 skipped
- # r6 = 0xffff_ffff_ffff_ff92 r3 is 1, so this is used
- # r7 = 0x0 skipped
-
+ """lst = ['sv.extsb/dm=1<<r3/sm=r30 5.v, 9.v']
+
+ one option for predicate masks is a single-bit set: 1<<r3.
+ lots of opportunity for hardware optimisation, it effectively
+ allows dynamic indexing of the register file
+
+ reg num 0 1 2 3 4 5 6 7 8 9 10 11
+ src r30=0b100 N N Y
+ |
+ +-----------+
+ |
+ dest r3=1: 1<<r3=0b010 N Y N
+
+ expected results:
+ r5 = 0x0 skipped
+ r6 = 0xffff_ffff_ffff_ff92 r3 is 1, so this is used
+ r7 = 0x0 skipped
+ """
isa = SVP64Asm(['sv.extsb/dm=1<<r3/sm=r30 5.v, 9.v'])
lst = list(isa)
print("listing", lst)
initial_svstate=svstate)
def case_17_shift_one_by_r3_source(self):
- # reg num 0 1 2 3 4 5 6 7 8 9 10 11
- # src r3=2: 1<<r3=0b100 N N Y
- # |
- # +-----------+
- # |
- # dest r30=0b010 N Y N
- #
- # expected results:
- # r5 = 0x0 skipped
- # r6 = 0xffff_ffff_ffff_ff92 2nd bit of r30 is 1
- # r7 = 0x0 skipped
-
+ """lst = ['sv.extsb/sm=1<<r3/dm=r30 5.v, 9.v']
+
+ reg num 0 1 2 3 4 5 6 7 8 9 10 11
+ src r3=2: 1<<r3=0b100 N N Y
+ |
+ +-----------+
+ |
+ dest r30=0b010 N Y N
+
+ expected results:
+ r5 = 0x0 skipped
+ r6 = 0xffff_ffff_ffff_ff92 2nd bit of r30 is 1
+ r7 = 0x0 skipped
+ """
isa = SVP64Asm(['sv.extsb/sm=1<<r3/dm=r30 5.v, 9.v'])
lst = list(isa)
print("listing", lst)
initial_svstate=svstate)
def case_18_sv_add_cr_pred(self):
- # adds, CR predicated mask CR4.eq = 1, CR5.eq = 0, invert (ne)
- # 1 = 5 + 9 => not to be touched (skipped)
- # 2 = 6 + 10 => 0x3334 = 0x2223+0x1111
- # expected results:
- # r1 = 0xbeef skipped since CR4 is 1 and test is inverted
- # r2 = 0x3334 CR5 is 0, so this is used
+ """lst = ['sv.add/m=ne 1.v, 5.v, 9.v']
+
+ adds, CR predicated mask CR4.eq = 1, CR5.eq = 0, invert (ne)
+ 1 = 5 + 9 => not to be touched (skipped)
+ 2 = 6 + 10 => 0x3334 = 0x2223+0x1111
+
+ expected results:
+ r1 = 0xbeef skipped since CR4 is 1 and test is inverted
+ r2 = 0x3334 CR5 is 0, so this is used
+ """
isa = SVP64Asm(['sv.add/m=ne 1.v, 5.v, 9.v'])
lst = list(isa)
print("listing", lst)
self.add_case(Program(lst, bigendian), initial_regs,
initial_svstate=svstate, initial_cr=cr)
- # checks reentrant CR predication
def case_19_crpred_reentrant(self):
- # reg num 0 1 2 3 4 5 6 7 8 9 10 11 12
- # srcstep=1 v
- # src cr4.eq=1 Y N Y N
- # cr6.eq=1 : |
- # + - - + |
- # : +-------+
- # dest cr5.lt=1 : |
- # cr7.lt=1 N Y N Y
- # dststep=2 ^
- #
- # expected results:
- # r5 = 0x0 skip
- # r6 = 0x0 dststep starts at 3, so this gets skipped
- # r7 = 0x0 skip
- # r8 = 0xffff_ffff_ffff_ff92 this will be used
-
+ """lst = ['sv.extsb/sm=eq/dm=lt 5.v, 9.v']
+
+ checks reentrant CR predication. note that the source CR-mask
+ and destination CR-mask use *different bits* of the CR fields,
+ despite both predicates starting from the same CR field number.
+ cr4.lt is zero, cr7.lt is zero AND
+ cr5.eq is zero, cr6.eq is zero.
+
+ reg num 0 1 2 3 4 5 6 7 8 9 10 11 12
+ srcstep=1 v
+ src cr4.eq=1 Y N Y N
+ cr6.eq=1 : |
+ + - - + |
+ : +-------+
+ dest cr5.lt=1 : |
+ cr7.lt=1 N Y N Y
+ dststep=2 ^
+
+ expected results:
+ r5 = 0x0 skip
+ r6 = 0x0 dststep starts at 3, so this gets skipped
+ r7 = 0x0 skip
+ r8 = 0xffff_ffff_ffff_ff92 this will be used
+ """
isa = SVP64Asm(['sv.extsb/sm=eq/dm=lt 5.v, 9.v'])
lst = list(isa)
print("listing", lst)
# set up CR predicate
# CR4.eq=1 and CR6.eq=1
cr.crl[4][CRFields.EQ] = 1
+ cr.crl[5][CRFields.EQ] = 0
cr.crl[6][CRFields.EQ] = 1
+ cr.crl[7][CRFields.EQ] = 0
# CR5.lt=1 and CR7.lt=1
+ cr.crl[4][CRFields.LT] = 0
cr.crl[5][CRFields.LT] = 1
+ cr.crl[6][CRFields.LT] = 0
cr.crl[7][CRFields.LT] = 1
# SVSTATE (in this case, VL=4)
svstate = SVP64State()