From a47842d16653b4f73b5d56ff0c252dd8a329481b Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Christoph=20M=C3=BCllner?= Date: Mon, 10 Jul 2023 09:12:43 +0200 Subject: riscv: Add support for the Zfa extension MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This patch introduces the RISC-V Zfa extension, which introduces additional floating-point instructions: * fli (load-immediate) with pre-defined immediates * fminm/fmaxm (like fmin/fmax but with different NaN behaviour) * fround/froundmx (round to integer) * fcvtmod.w.d (Modular Convert-to-Integer) * fmv* to access high bits of float register bigger than XLEN * Quiet comparison instructions (fleq/fltq) Zfa defines its instructions in combination with the following extensions: * single-precision floating-point (F) * double-precision floating-point (D) * quad-precision floating-point (Q) * half-precision floating-point (Zfh) Since QEMU does not support the RISC-V quad-precision floating-point ISA extension (Q), this patch does not include the instructions that depend on this extension. All other instructions are included in this patch. The Zfa specification can be found here: https://github.com/riscv/riscv-isa-manual/blob/master/src/zfa.tex The Zfa specifciation is frozen and is in public review since May 3, 2023: https://groups.google.com/a/groups.riscv.org/g/isa-dev/c/SED4ntBkabg The patch also includes a TCG test for the fcvtmod.w.d instruction. The test cases test for correct results and flag behaviour. Note, that the Zfa specification requires fcvtmod's flag behaviour to be identical to a fcvt with the same operands (which is also tested). Reviewed-by: Richard Henderson Signed-off-by: Christoph Müllner Message-Id: <20230710071243.282464-1-christoph.muellner@vrull.eu> Signed-off-by: Alistair Francis --- tests/tcg/riscv64/Makefile.target | 6 + tests/tcg/riscv64/test-fcvtmod.c | 345 ++++++++++++++++++++++++++++++++++++++ 2 files changed, 351 insertions(+) create mode 100644 tests/tcg/riscv64/test-fcvtmod.c (limited to 'tests/tcg') diff --git a/tests/tcg/riscv64/Makefile.target b/tests/tcg/riscv64/Makefile.target index 42993549cb..a7e390c384 100644 --- a/tests/tcg/riscv64/Makefile.target +++ b/tests/tcg/riscv64/Makefile.target @@ -12,3 +12,9 @@ run-test-noc: QEMU_OPTS += -cpu rv64,c=false TESTS += test-aes run-test-aes: QEMU_OPTS += -cpu rv64,zk=on + +# Test for fcvtmod +TESTS += test-fcvtmod +test-fcvtmod: CFLAGS += -march=rv64imafdc +test-fcvtmod: LDFLAGS += -static +run-test-fcvtmod: QEMU_OPTS += -cpu rv64,d=true,Zfa=true diff --git a/tests/tcg/riscv64/test-fcvtmod.c b/tests/tcg/riscv64/test-fcvtmod.c new file mode 100644 index 0000000000..f050579974 --- /dev/null +++ b/tests/tcg/riscv64/test-fcvtmod.c @@ -0,0 +1,345 @@ +#include +#include +#include + +#define FFLAG_NX_SHIFT 0 /* inexact */ +#define FFLAG_UF_SHIFT 1 /* underflow */ +#define FFLAG_OF_SHIFT 2 /* overflow */ +#define FFLAG_DZ_SHIFT 3 /* divide by zero */ +#define FFLAG_NV_SHIFT 4 /* invalid operation */ + +#define FFLAG_NV (1UL << FFLAG_NV_SHIFT) +#define FFLAG_DZ (1UL << FFLAG_DZ_SHIFT) +#define FFLAG_OF (1UL << FFLAG_OF_SHIFT) +#define FFLAG_UF (1UL << FFLAG_UF_SHIFT) +#define FFLAG_NX (1UL << FFLAG_NX_SHIFT) + +typedef struct fp64_fcvt_fcvtmod_testcase { + const char* name; + union { + uint64_t inp_lu; + double inp_lf; + }; + uint64_t exp_fcvt; + uint8_t exp_fcvt_fflags; + uint64_t exp_fcvtmod; + uint8_t exp_fcvtmod_fflags; +} fp64_fcvt_fcvtmod_testcase_t; + +void print_fflags(uint8_t fflags) +{ + int set = 0; + + if (fflags == 0) { + printf("-"); + return; + } + + if (fflags & FFLAG_NV) { + printf("%sFFLAG_NV", set ? " | " : ""); + set = 1; + } + if (fflags & FFLAG_DZ) { + printf("%sFFLAG_DZ", set ? " | " : ""); + set = 1; + } + if (fflags & FFLAG_OF) { + printf("%sFFLAG_OF", set ? " | " : ""); + set = 1; + } + if (fflags & FFLAG_UF) { + printf("%sFFLAG_UF", set ? " | " : ""); + set = 1; + } + if (fflags & FFLAG_NX) { + printf("%sFFLAG_NX", set ? " | " : ""); + set = 1; + } +} + +/* Clear all FP flags. */ +static inline void clear_fflags() +{ + __asm__ __volatile__("fsflags zero"); +} + +/* Read all FP flags. */ +static inline uint8_t get_fflags() +{ + uint64_t v; + __asm__ __volatile__("frflags %0" : "=r"(v)); + return (uint8_t)v; +} + +/* Move input value (without conversations) into an FP register. */ +static inline double do_fmv_d_x(uint64_t inp) +{ + double fpr; + __asm__ __volatile__("fmv.d.x %0, %1" : "=f"(fpr) : "r"(inp)); + return fpr; +} + +static inline uint64_t do_fcvt_w_d(uint64_t inp, uint8_t *fflags) +{ + uint64_t ret; + double fpr = do_fmv_d_x(inp); + + clear_fflags(); + + __asm__ __volatile__("fcvt.w.d %0, %1, rtz" : "=r"(ret) : "f"(fpr)); + + *fflags = get_fflags(); + + return ret; +} + +static inline uint64_t do_fcvtmod_w_d(uint64_t inp, uint8_t *fflags) +{ + uint64_t ret; + double fpr = do_fmv_d_x(inp); + + clear_fflags(); + + /* fcvtmod.w.d rd, rs1, rtz = 1100001 01000 rs1 001 rd 1010011 */ + asm(".insn r 0x53, 0x1, 0x61, %0, %1, f8" : "=r"(ret) : "f"(fpr)); + + *fflags = get_fflags(); + + return ret; +} + +static const fp64_fcvt_fcvtmod_testcase_t tests[] = { + /* Zero (exp=0, frac=0) */ + { .name = "+0.0", + .inp_lf = 0x0p0, + .exp_fcvt = 0x0000000000000000, + .exp_fcvt_fflags = 0, + .exp_fcvtmod = 0x0000000000000000, + .exp_fcvtmod_fflags = 0 }, + { .name = "-0.0", + .inp_lf = -0x0p0, + .exp_fcvt = 0x0000000000000000, + .exp_fcvt_fflags = 0, + .exp_fcvtmod = 0x0000000000000000, + .exp_fcvtmod_fflags = 0 }, + + /* Subnormal: exp=0 frac!=0 */ + { .name = "Subnormal frac=1", + .inp_lu = 0x0000000000000001, + .exp_fcvt = 0x0000000000000000, + .exp_fcvt_fflags = FFLAG_NX, + .exp_fcvtmod = 0, + .exp_fcvtmod_fflags = FFLAG_NX }, + { .name = "Subnormal frac=0xf..f", + .inp_lu = 0x0000ffffffffffff, + .exp_fcvt = 0x0000000000000000, + .exp_fcvt_fflags = FFLAG_NX, + .exp_fcvtmod = 0, + .exp_fcvtmod_fflags = FFLAG_NX }, + { .name = "Neg subnormal frac=1", + .inp_lu = 0x0000000000000001, + .exp_fcvt = 0x0000000000000000, + .exp_fcvt_fflags = FFLAG_NX, + .exp_fcvtmod = 0, + .exp_fcvtmod_fflags = FFLAG_NX }, + { .name = "Neg subnormal frac=0xf..f", + .inp_lu = 0x8000ffffffffffff, + .exp_fcvt = 0x0000000000000000, + .exp_fcvt_fflags = FFLAG_NX, + .exp_fcvtmod = 0, + .exp_fcvtmod_fflags = FFLAG_NX }, + + /* Infinity: exp=0x7ff, frac=0 */ + { .name = "+INF", + .inp_lu = 0x7ff0000000000000, + .exp_fcvt = 0x000000007fffffff, /* int32 max */ + .exp_fcvt_fflags = FFLAG_NV, + .exp_fcvtmod = 0, + .exp_fcvtmod_fflags = FFLAG_NV }, + { .name = "-INF", + .inp_lu = 0xfff0000000000000, + .exp_fcvt = 0xffffffff80000000, /* int32 min */ + .exp_fcvt_fflags = FFLAG_NV, + .exp_fcvtmod = 0, + .exp_fcvtmod_fflags = FFLAG_NV }, + + /* NaN: exp=7ff, frac!=0 */ + { .name = "canonical NaN", + .inp_lu = 0x7ff8000000000000, + .exp_fcvt = 0x000000007fffffff, /* int32 max */ + .exp_fcvt_fflags = FFLAG_NV, + .exp_fcvtmod = 0, + .exp_fcvtmod_fflags = FFLAG_NV }, + { .name = "non-canonical NaN", + .inp_lu = 0x7ff8000000100000, + .exp_fcvt = 0x000000007fffffff, /* int32 min */ + .exp_fcvt_fflags = FFLAG_NV, + .exp_fcvtmod = 0, + .exp_fcvtmod_fflags = FFLAG_NV }, + + /* Normal numbers: exp!=0, exp!=7ff */ + { .name = "+smallest normal value", + .inp_lu = 0x0010000000000000, + .exp_fcvt = 0, + .exp_fcvt_fflags = FFLAG_NX, + .exp_fcvtmod = 0, + .exp_fcvtmod_fflags = FFLAG_NX }, + { .name = "-smallest normal value", + .inp_lu = 0x8010000000000000, + .exp_fcvt = 0, + .exp_fcvt_fflags = FFLAG_NX, + .exp_fcvtmod = 0, + .exp_fcvtmod_fflags = FFLAG_NX }, + + { .name = "+0.5", + .inp_lf = 0x1p-1, + .exp_fcvt = 0, + .exp_fcvt_fflags = FFLAG_NX, + .exp_fcvtmod = 0, + .exp_fcvtmod_fflags = FFLAG_NX }, + { .name = "-0.5", + .inp_lf = -0x1p-1, + .exp_fcvt = 0, + .exp_fcvt_fflags = FFLAG_NX, + .exp_fcvtmod = 0, + .exp_fcvtmod_fflags = FFLAG_NX }, + + { .name = "+value just below 1.0", + .inp_lu = 0x3fefffffffffffff, + .exp_fcvt = 0, + .exp_fcvt_fflags = FFLAG_NX, + .exp_fcvtmod = 0, + .exp_fcvtmod_fflags = FFLAG_NX }, + { .name = "-value just above -1.0", + .inp_lu = 0xbfefffffffffffff, + .exp_fcvt = 0, + .exp_fcvt_fflags = FFLAG_NX, + .exp_fcvtmod = 0, + .exp_fcvtmod_fflags = FFLAG_NX }, + + { .name = "+1.0", + .inp_lf = 0x1p0, + .exp_fcvt = 0x0000000000000001, + .exp_fcvt_fflags = 0, + .exp_fcvtmod = 0x0000000000000001, + .exp_fcvtmod_fflags = 0 }, + { .name = "-1.0", + .inp_lf = -0x1p0, + .exp_fcvt = 0xffffffffffffffff, + .exp_fcvt_fflags = 0, + .exp_fcvtmod = 0xffffffffffffffff, + .exp_fcvtmod_fflags = 0 }, + + { .name = "+1.5", + .inp_lu = 0x3ff8000000000000, + .exp_fcvt = 1, + .exp_fcvt_fflags = FFLAG_NX, + .exp_fcvtmod = 1, + .exp_fcvtmod_fflags = FFLAG_NX }, + { .name = "-1.5", + .inp_lu = 0xbff8000000000000, + .exp_fcvt = 0xffffffffffffffff, + .exp_fcvt_fflags = FFLAG_NX, + .exp_fcvtmod = 0xffffffffffffffff, + .exp_fcvtmod_fflags = FFLAG_NX }, + + { .name = "+max int32 (2147483647)", + .inp_lu = 0x41dfffffffc00000, + .exp_fcvt = 0x000000007fffffff, + .exp_fcvt_fflags = 0, + .exp_fcvtmod = 0x000000007fffffff, + .exp_fcvtmod_fflags = 0 }, + { .name = "+max int32 +1 (2147483648)", + .inp_lf = 0x1p31, + .exp_fcvt = 0x000000007fffffff, + .exp_fcvt_fflags = FFLAG_NV, + .exp_fcvtmod = (uint64_t)-2147483648l, /* int32 min */ + .exp_fcvtmod_fflags = FFLAG_NV }, + { .name = "+max int32 +2 (2147483649)", + .inp_lu = 0x41e0000000200000, + .exp_fcvt = 0x000000007fffffff, + .exp_fcvt_fflags = FFLAG_NV, + .exp_fcvtmod = (uint64_t)-2147483647l, /* int32 min +1 */ + .exp_fcvtmod_fflags = FFLAG_NV }, + + { .name = "-max int32 (-2147483648)", + .inp_lf = -0x1p31, + .exp_fcvt = 0xffffffff80000000, + .exp_fcvt_fflags = 0, + .exp_fcvtmod = 0xffffffff80000000, + .exp_fcvtmod_fflags = 0 }, + { .name = "-max int32 -1 (-2147483649)", + .inp_lf = -0x1.00000002p+31, + .exp_fcvt = 0xffffffff80000000, + .exp_fcvt_fflags = FFLAG_NV, + .exp_fcvtmod = 2147483647, /* int32 max */ + .exp_fcvtmod_fflags = FFLAG_NV }, + { .name = "-max int32 -2 (-2147483650)", + .inp_lf = -0x1.00000004p+31, + .exp_fcvt = 0xffffffff80000000, + .exp_fcvt_fflags = FFLAG_NV, + .exp_fcvtmod = 2147483646, /* int32 max -1 */ + .exp_fcvtmod_fflags = FFLAG_NV }, +}; + +int run_fcvtmod_tests() +{ + uint64_t act_fcvt; + uint8_t act_fcvt_fflags; + uint64_t act_fcvtmod; + uint8_t act_fcvtmod_fflags; + + for (size_t i = 0; i < sizeof(tests)/sizeof(tests[0]); i++) { + const fp64_fcvt_fcvtmod_testcase_t *t = &tests[i]; + + act_fcvt = do_fcvt_w_d(t->inp_lu, &act_fcvt_fflags); + int fcvt_correct = act_fcvt == t->exp_fcvt && + act_fcvt_fflags == t->exp_fcvt_fflags; + act_fcvtmod = do_fcvtmod_w_d(t->inp_lu, &act_fcvtmod_fflags); + int fcvtmod_correct = act_fcvtmod == t->exp_fcvtmod && + act_fcvtmod_fflags == t->exp_fcvtmod_fflags; + + if (fcvt_correct && fcvtmod_correct) { + continue; + } + + printf("Test %zu (%s) failed!\n", i, t->name); + + double fpr = do_fmv_d_x(t->inp_lu); + printf("inp_lu: 0x%016lx == %lf\n", t->inp_lu, fpr); + printf("inp_lf: %lf\n", t->inp_lf); + + uint32_t sign = (t->inp_lu >> 63); + uint32_t exp = (uint32_t)(t->inp_lu >> 52) & 0x7ff; + uint64_t frac = t->inp_lu & 0xfffffffffffffull; /* significand */ + int true_exp = exp - 1023; + int shift = true_exp - 52; + uint64_t true_frac = frac | 1ull << 52; + + printf("sign=%d, exp=0x%03x, frac=0x%012lx\n", sign, exp, frac); + printf("true_exp=%d, shift=%d, true_frac=0x%016lx\n", true_exp, shift, true_frac); + + if (!fcvt_correct) { + printf("act_fcvt: 0x%016lx == %li\n", act_fcvt, act_fcvt); + printf("exp_fcvt: 0x%016lx == %li\n", t->exp_fcvt, t->exp_fcvt); + printf("act_fcvt_fflags: "); print_fflags(act_fcvt_fflags); printf("\n"); + printf("exp_fcvt_fflags: "); print_fflags(t->exp_fcvt_fflags); printf("\n"); + } + + if (!fcvtmod_correct) { + printf("act_fcvtmod: 0x%016lx == %li\n", act_fcvtmod, act_fcvtmod); + printf("exp_fcvtmod: 0x%016lx == %li\n", t->exp_fcvtmod, t->exp_fcvtmod); + printf("act_fcvtmod_fflags: "); print_fflags(act_fcvtmod_fflags); printf("\n"); + printf("exp_fcvtmod_fflags: "); print_fflags(t->exp_fcvtmod_fflags); printf("\n"); + } + + return 1; + } + + return 0; +} + +int main() +{ + return run_fcvtmod_tests(); +} -- cgit v1.2.3