From: Luke Kenneth Casson Leighton Date: Sun, 11 Jul 2021 23:02:00 +0000 (+0100) Subject: minor reordering of setvl and svshape: svshape is now capable of setting VL X-Git-Tag: xlen-bcd~293 X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=0b7e02cfc226e031a9cac3118acc52e9c29a0336;p=openpower-isa.git minor reordering of setvl and svshape: svshape is now capable of setting VL --- diff --git a/media/audio/mp3/mp3_0_apply_window_float_basicsv.s b/media/audio/mp3/mp3_0_apply_window_float_basicsv.s index f8f0b283..ebbdedb2 100644 --- a/media/audio/mp3/mp3_0_apply_window_float_basicsv.s +++ b/media/audio/mp3/mp3_0_apply_window_float_basicsv.s @@ -53,7 +53,7 @@ ff_mpadsp_apply_window_float_sv: add out2, out, 0 # set Vector Length - setvl 0, 0, 7, 0, 1, 1 # setvli MVL=8, VL=8 + setvl 0, 0, 7, 1, 1, 0# setvli MVL=8, VL=8 addi win2, win, 124 # w2 = window + 31 lfiwax sum, 0, 5 # sum = *dither_state diff --git a/openpower/isa/simplev.mdwn b/openpower/isa/simplev.mdwn index 667d1dad..0723a2d7 100644 --- a/openpower/isa/simplev.mdwn +++ b/openpower/isa/simplev.mdwn @@ -31,7 +31,7 @@ Pseudo-code: SVSTATE[0:6] <- MVL SVSTATE[7:13] <- VL if _RT != 0b00000 then - GPR(_RT) <- [0]*57 || VL + GPR(_RT) <- [0]*57 || VL MSR[6] <- vf Special Registers Altered: @@ -62,17 +62,23 @@ Special Registers Altered: SVM-Form -* svshape SVxd, SVyd, SVzd, SVRM +* svshape SVxd, SVyd, SVzd, SVRM, vf Pseudo-code: + # for convenience, VL to be calculated and stored in SVSTATE + vlen <- [0] * 7 + SVSTATE[0:31] <- [0] * 32 # clear out all SVSHAPEs SVSHAPE0[0:31] <- [0] * 32 SVSHAPE1[0:31] <- [0] * 32 SVSHAPE2[0:31] <- [0] * 32 SVSHAPE3[0:31] <- [0] * 32 # set schedule up for multiply - if (SVRM = 0b00000) then + if (SVRM = 0b0000) then + # VL in Matrix Multiply is xd*yd*zd + n <- (0b00 || SVxd) * (0b00 || SVyd) * (0b00 || SVzd) + vlen[0:6] <- n[14:20] # set up template in SVSHAPE0, then copy to 1-3 SVSHAPE0[0:5] <- (0b0 || SVxd) # xdim SVSHAPE0[6:11] <- (0b0 || SVyd) # ydim @@ -89,7 +95,15 @@ Pseudo-code: SVSHAPE2[18:20] <- 0b001 # permute x,z,y SVSHAPE2[28:29] <- 0b11 # skip y # set schedule up for butterfly - if (SVRM = 0b00001) then + if (SVRM = 0b0001) then + # calculate O(N log2 N) + n <- [0] * 3 + do while n < 5 + if SVxd[4-n] = 0 then + leave + n <- n + 1 + n <- ((0b0 || SVxd) + 1) * n + vlen[0:6] <- n[1:7] # set up template in SVSHAPE0, then copy to 1-3 # for FRA and FRT SVSHAPE0[0:5] <- (0b0 || SVxd) # xdim @@ -101,6 +115,10 @@ Pseudo-code: SVSHAPE1[28:29] <- 0b01 # j+halfstep schedule # FRC (coefficients) SVSHAPE2[28:29] <- 0b10 # k schedule + # set VL, MVL and MSR Vertical-First + SVSTATE[0:6] <- vlen + SVSTATE[7:13] <- vlen + MSR[6] <- vf Special Registers Altered: diff --git a/openpower/isatables/fields.text b/openpower/isatables/fields.text index 511e91c7..5c2bb828 100644 --- a/openpower/isatables/fields.text +++ b/openpower/isatables/fields.text @@ -253,7 +253,7 @@ # 1.6.28 SVL-FORM |0 |6 |11 |16 |23 |24 |25 |26 |31 | - | PO | RT | RA | SVi |vf |vs |ms | XO |Rc | + | PO | RT | RA | SVi |ms |vs |vf | XO |Rc | # 1.6.29 SVC-FORM |0 |6 |9 |11 | @@ -276,8 +276,8 @@ | PO | RS | RA | RC | SVDS | XO | # 1.6.33 SVM-FORM - |0 |6 |11 |16 |21 |26 |31 | - | PO | SVxd | SVyd | SVzd | SVRM | XO | / | + |0 |6 |11 |16 |21 |25 |26 |31 | + | PO | SVxd | SVyd | SVzd | SVRM |vf | XO | / | # 1.6.34 SVRM-FORM |0 |6 |11 |13 |15 |17 |19 |21 |26 |31 | @@ -607,7 +607,7 @@ Field used in X-form instructions to specify a sub- set of storage accesses. Formats: X - ms (25) + ms (23) Field used in Simple-V to specify whether MVL is to be set Formats: SVL NB (16:20) @@ -783,7 +783,7 @@ SVme (6:10) Simple-V "REMAP" map-enable bits (0-4) Formats: SVRM - SVRM (21:25) + SVRM (21:24) Simple-V "REMAP" Mode Formats: SVM SVxd (6:10) @@ -877,9 +877,9 @@ VRT (6:10) Field used to specify a VR to be used as a target. Formats: DS, VA, VC, VX, X - vf (23) + vf (25) Field used in Simple-V to specify whether "Vertical" Mode is set - Formats: SVL + Formats: SVL, SVM vs (24) Field used in Simple-V to specify whether VL is to be set Formats: SVL diff --git a/src/openpower/decoder/isa/test_caller_svp64_fft.py b/src/openpower/decoder/isa/test_caller_svp64_fft.py index c7854ee8..d7f9ae9a 100644 --- a/src/openpower/decoder/isa/test_caller_svp64_fft.py +++ b/src/openpower/decoder/isa/test_caller_svp64_fft.py @@ -122,7 +122,7 @@ class FFTTestCase(FHDLTestCase): self.assertEqual(sim.gpr(i), SelectableInt(expected[i], 64)) def test_sv_remap_fpmadds_fft(self): - """>>> lst = ["svshape 8, 1, 1, 1", + """>>> lst = ["svshape 8, 1, 1, 1, 0", "svremap 31, 1, 0, 2, 0, 1", "sv.ffmadds 2.v, 2.v, 2.v, 10.v" ] @@ -138,7 +138,7 @@ class FFTTestCase(FHDLTestCase): SVP64 "REMAP" in Butterfly Mode is applied to a twin +/- FMAC (3 inputs, 2 outputs) """ - lst = SVP64Asm( ["svshape 8, 1, 1, 1", + lst = SVP64Asm( ["svshape 8, 1, 1, 1, 0", "svremap 31, 1, 0, 2, 0, 1", "sv.ffmadds 0.v, 0.v, 0.v, 8.v" ]) @@ -156,28 +156,8 @@ class FFTTestCase(FHDLTestCase): for i, a in enumerate(av): fprs[i+0] = fp64toselectable(a) - # set total. err don't know how to calculate how many there are... - # do it manually for now - VL = 0 - size = 2 - n = len(av) - while size <= n: - halfsize = size // 2 - tablestep = n // size - for i in range(0, n, size): - for j in range(i, i + halfsize): - VL += 1 - size *= 2 - - # SVSTATE (calculated VL) - svstate = SVP64State() - svstate.vl[0:7] = VL # VL - svstate.maxvl[0:7] = VL # MAXVL - print ("SVSTATE", bin(svstate.spr.asint())) - with Program(lst, bigendian=False) as program: - sim = self.run_tst_program(program, svstate=svstate, - initial_fprs=fprs) + sim = self.run_tst_program(program, initial_fprs=fprs) print ("spr svshape0", sim.spr['SVSHAPE0']) print (" xdimsz", sim.spr['SVSHAPE0'].xdimsz) print (" ydimsz", sim.spr['SVSHAPE0'].ydimsz) @@ -203,8 +183,8 @@ class FFTTestCase(FHDLTestCase): self.assertTrue(err < 1e-7) def test_sv_remap_fpmadds_fft_svstep(self): - """>>> lst = SVP64Asm( ["setvl 0, 0, 11, 1, 1, 1", - "svshape 8, 1, 1, 1", + """>>> lst = SVP64Asm( [ + "svshape 8, 1, 1, 1, 1", "svremap 31, 1, 0, 2, 0, 1", "sv.ffmadds 0.v, 0.v, 0.v, 8.v", "setvl. 0, 0, 0, 1, 0, 0", @@ -218,8 +198,8 @@ class FFTTestCase(FHDLTestCase): SVP64 "REMAP" in Butterfly Mode is applied to a twin +/- FMAC (3 inputs, 2 outputs) """ - lst = SVP64Asm( ["setvl 0, 0, 11, 1, 1, 1", - "svshape 8, 1, 1, 1", + lst = SVP64Asm( [ + "svshape 8, 1, 1, 1, 1", "svremap 31, 1, 0, 2, 0, 1", "sv.ffmadds 0.v, 0.v, 0.v, 8.v", "setvl. 0, 0, 0, 1, 0, 0", @@ -416,18 +396,17 @@ class FFTTestCase(FHDLTestCase): vec[jl] = temp2 + temp1 """ lst = SVP64Asm( ["setvl 0, 0, 11, 1, 1, 1", - "svshape 8, 1, 1, 1", + "svshape 8, 1, 1, 1, 1", # tpre "svremap 31, 1, 0, 2, 0, 1", "sv.fmuls 24, 0.v, 16.v", - "svshape 8, 1, 1, 1", "svremap 31, 1, 0, 2, 0, 1", "sv.fmuls 25, 8.v, 20.v", "fadds 24, 24, 25", # tpim "svremap 31, 1, 0, 2, 0, 1", "sv.fmuls 26, 0.v, 20.v", - "svshape 8, 1, 1, 1", + "svremap 31, 1, 0, 2, 0, 1", "sv.fmuls 26, 8.v, 16.v", "fsubs 26, 26, 27", # vec_r jh/jl diff --git a/src/openpower/decoder/isa/test_caller_svp64_matrix.py b/src/openpower/decoder/isa/test_caller_svp64_matrix.py index 4f5519af..99ca74df 100644 --- a/src/openpower/decoder/isa/test_caller_svp64_matrix.py +++ b/src/openpower/decoder/isa/test_caller_svp64_matrix.py @@ -27,13 +27,13 @@ class DecoderTestCase(FHDLTestCase): self.assertEqual(sim.gpr(i), SelectableInt(expected[i], 64)) def test_sv_remap1(self): - """>>> lst = ["svshape 2, 2, 3, 0", + """>>> lst = ["svshape 2, 2, 3, 0, 0", "svremap 31, 1, 2, 3, 0, 0", "sv.fmadds 0.v, 8.v, 16.v, 0.v" ] REMAP fmadds FRT, FRA, FRC, FRB """ - lst = SVP64Asm(["svshape 2, 2, 3, 0", + lst = SVP64Asm(["svshape 2, 2, 3, 0, 0", "svremap 31, 1, 2, 3, 0, 0", "sv.fmadds 0.v, 16.v, 32.v, 0.v" ]) @@ -100,13 +100,13 @@ class DecoderTestCase(FHDLTestCase): # self.assertEqual(sim.fpr(i+6), u) def test_sv_remap2(self): - """>>> lst = ["svshape 5, 4, 3, 0", + """>>> lst = ["svshape 5, 4, 3, 0, 0", "svremap 31, 1, 2, 3, 0, 0", "sv.fmadds 0.v, 8.v, 16.v, 0.v" ] REMAP fmadds FRT, FRA, FRC, FRB """ - lst = SVP64Asm(["svshape 4, 3, 3, 0", + lst = SVP64Asm(["svshape 4, 3, 3, 0, 0", "svremap 31, 1, 2, 3, 0, 0", "sv.fmadds 0.v, 16.v, 32.v, 0.v" ]) @@ -153,7 +153,6 @@ class DecoderTestCase(FHDLTestCase): xdim2 = len(Y[0]) ydim2 = len(Y) - VL = ydim2 * xdim2 * ydim1 print ("xdim2 ydim1 ydim2", xdim2, ydim1, ydim2) xf = reduce(operator.add, X) @@ -180,15 +179,8 @@ class DecoderTestCase(FHDLTestCase): #print ("FFT", i, "in", a, b, "coeff", c, "mul", # mul, "res", t, u) - # SVSTATE (in this case, VL=12, to cover all of matrix) - svstate = SVP64State() - svstate.vl[0:7] = VL # VL - svstate.maxvl[0:7] = VL # MAXVL - print ("SVSTATE", bin(svstate.spr.asint())) - with Program(lst, bigendian=False) as program: - sim = self.run_tst_program(program, svstate=svstate, - initial_fprs=fprs) + sim = self.run_tst_program(program, initial_fprs=fprs) print ("spr svshape0", sim.spr['SVSHAPE0']) print (" xdimsz", sim.spr['SVSHAPE0'].xdimsz) print (" ydimsz", sim.spr['SVSHAPE0'].ydimsz) diff --git a/src/openpower/sv/trans/svp64.py b/src/openpower/sv/trans/svp64.py index 4216bed7..05c768fd 100644 --- a/src/openpower/sv/trans/svp64.py +++ b/src/openpower/sv/trans/svp64.py @@ -187,9 +187,9 @@ class SVP64Asm: insn |= fields[0] << (31-10) # RT , bits 6-10 insn |= fields[1] << (31-15) # RA , bits 11-15 insn |= fields[2] << (31-22) # SVi , bits 16-22 - insn |= fields[3] << (31-23) # vf , bit 23 + insn |= fields[3] << (31-25) # ms , bit 25 insn |= fields[4] << (31-24) # vs , bit 24 - insn |= fields[5] << (31-25) # ms , bit 25 + insn |= fields[5] << (31-23) # vf , bit 23 insn |= 0b00000 << (31-30) # XO , bits 26..30 if opcode == 'setvl.': insn |= 1 << (31-31) # Rc=1 , bit 31 @@ -204,7 +204,8 @@ class SVP64Asm: insn |= (fields[0]-1) << (31-10) # SVxd , bits 6-10 insn |= (fields[1]-1) << (31-15) # SVyd , bits 11-15 insn |= (fields[2]-1) << (31-20) # SVzd , bits 16-20 - insn |= (fields[3]) << (31-25) # SVRM , bits 21-25 + insn |= (fields[3]) << (31-24) # SVRM , bits 21-24 + insn |= (fields[4]) << (31-25) # vf , bits 25 insn |= 0b00001 << (31-30) # XO , bits 26..30 #insn &= ((1<<32)-1) log ("svshape", bin(insn)) @@ -998,9 +999,10 @@ if __name__ == '__main__': ] lst = [ #'sv.fmadds 0.v, 8.v, 16.v, 4.v', - #'svshape 8, 1, 1, 1', #'sv.ffadds 0.v, 8.v, 4.v', - 'svremap 11, 0, 1, 2, 3, 2', + #'svremap 11, 0, 1, 2, 3, 2', + 'svshape 8, 1, 1, 1, 0', + 'svshape 8, 1, 1, 1, 1', ] isa = SVP64Asm(lst, macros=macros) print ("list", list(isa))