-march may be wrong when using pulpissimo configuration

[hongzhi2015]

When running tests from https://github.com/pulp-platform/regression_tests/

At least, I found 3 cases freeze forever:

• sequential_bare_tests/seidel
• sequential_bare_tests/fdtd-1d
• sequential_bare_tests/jacobi-1d-imper

And I use sequential_bare_tests/seidel to debug, the simulation hangs just before floating-point operations:

        a[i][j] = (a[i-1][j-1] + a[i-1][j] + a[i-1][j+1]
            + a[i][j-1] + a[i][j] + a[i][j+1]
            + a[i+1][j-1] + a[i+1][j] + a[i+1][j+1])/9.0f;

Finally, I edited rules/pulpos/targets/pulpissimo.mk, use -march=rv32imfcxpulpv2, and all the 3 cases run to the end.

Original

ifdef USE_IBEX
PULP_LDFLAGS +=
PULP_CFLAGS += -D__ibex__ -U__riscv__ -UARCHI_CORE_HAS_PULPV2 -DRV_ISA_RV32
PULP_ARCH_CFLAGS ?= -march=rv32imc
PULP_ARCH_LDFLAGS ?= -march=rv32imc
PULP_ARCH_OBJDFLAGS ?= -Mmarch=rv32imc
else
PULP_LDFLAGS      += 
PULP_CFLAGS       +=  -D__riscv__
# PULP_ARCH_CFLAGS ?=  -march=rv32imcxgap9
# PULP_ARCH_LDFLAGS ?=  -march=rv32imcxgap9
# PULP_ARCH_OBJDFLAGS ?= -Mmarch=rv32imcxgap9
PULP_ARCH_CFLAGS ?=  -march=rv32imfcxpulpv2
PULP_ARCH_LDFLAGS ?=  -march=rv32imfcxpulpv2
PULP_ARCH_OBJDFLAGS ?= -Mmarch=rv32imfcxpulpv2
endif

I tried to use -march=rv32imfcxvpulpv3 as pulp-platform/pulpissimo#183 , and encounter the same error:

cc1: error: -march=rv32imfcxvpulpv3: unsupported ISA substring XVPULPV3

I checked pulp-sdk build by cd pulpissimo; make build-pulp-sdk, -march=rv32imfcxpulpv2 is used for pulpissimo, -march=rv32imcxgap9 is used for vega.

[hongzhi2015]

More information.
When using -march=rv32imcxgap9, wait for the simulation hang and stop the simulation, last lines of regression_tests/sequential_bare_tests/seidel/build/trace_core_1f_0.log are:

...
8553554ns          141043 1c0081e6 0005a703 lw               x14, 0(x11)         x14=00000000 x11:1c010000  PA:1c010000 
8553574ns          141044 1c0081e8 0045a783 lw               x15, 4(x11)         x15=3f800000 x11:1c010000  PA:1c010004 
8553594ns          141045 1c0081ea 0085a503 lw               x10, 8(x11)         x10=40800000 x11:1c010000  PA:1c010008 
8553614ns          141046 1c0081ec 0145ae03 lw               x28, 20(x11)        x28=3f800000 x11:1c010000  PA:1c010014 
8553634ns          141047 1c0081f0 00f77753 fadd.s           f14, f14, f15       f14=xxxxxxxx f14:xxxxxxxx f15:xxxxxxxx 
8553653ns          141048 1c0081f4 0185a783 lw               x15, 24(x11)        x15=40000000 x11:1c010000  PA:1c010018 
8553673ns          141049 1c0081f6 01c5a303 lw               x6, 28(x11)          x6=40a00000 x11:1c010000  PA:1c01001c 
8553693ns          141050 1c0081fa 0285a883 lw               x17, 40(x11)        x17=40800000 x11:1c010000  PA:1c010028 
8553713ns          141051 1c0081fe 02c5a803 lw               x16, 44(x11)        x16=40a00000 x11:1c010000  PA:1c01002c 
8553733ns          141052 1c008202 00a77753 fadd.s           f14, f14, f10       f14=xxxxxxxx f14:xxxxxxxx f10:xxxxxxxx 
8553752ns          141053 1c008206 0305a503 lw               x10, 48(x11)        x10=41000000 x11:1c010000  PA:1c010030 
8553772ns          141054 1c008208 00160613 addi             x12, x12, 1         x12=00000001 x12:00000000 
8553792ns          141055 1c00820a 01c77753 fadd.s           f14, f14, f28       f14=xxxxxxxx f14:xxxxxxxx f28:xxxxxxxx 
8553812ns          141056 1c00820e 00f777d3 fadd.s           f15, f14, f15       f15=xxxxxxxx f14:xxxxxxxx f15:xxxxxxxx 
8553831ns          141057 1c008212 0067f7d3 fadd.s           f15, f15, f6        f15=xxxxxxxx f15:xxxxxxxx  f6:xxxxxxxx 
8553851ns          141058 1c008216 0117f7d3 fadd.s           f15, f15, f17       f15=xxxxxxxx f15:xxxxxxxx f17:xxxxxxxx 
8553871ns          141059 1c00821a 0107f7d3 fadd.s           f15, f15, f16       f15=xxxxxxxx f15:xxxxxxxx f16:xxxxxxxx 

Floating point registers contains 'x'.