Zvfqwdota8f: dot product of 8-bit floating-point numbers with 32-bit accumulation.
Zvdota also uses the vtype alternative format bit, altfmt, which is bit 8 of the vtype CSR. This patch adds symbolic e8alt and e16alt vsew constants for vsetvli/vsetivli so that assembly code can select the alternative element formats required by these dot-product instructions.
[1] https://github.com/riscv/riscv-isa-manual/blob/zvbdot/src/zvdota.adoc
bfd/ChangeLog:
* elfxx-riscv.c (riscv_implicit_subsets): New implies.
(riscv_supported_std_z_ext): New extensions.
(riscv_multi_subset_supports): Ditto.
(riscv_multi_subset_supports_ext): Ditto.
gas/ChangeLog:
* NEWS: Add zvqwdota8i, zvqwdota16i, zvfwdota16bf and
zvfqwdota8f extensions.
* config/tc-riscv.c (riscv_vsew_altfmt): New vsew altfmt
constants array for e8alt and e16alt.
(riscv_vtype_altfmt_supported): New function.
(my_getVsetvliExpression): Ditto.
* testsuite/gas/riscv/march-help.l: New test.
* testsuite/gas/riscv/vector-vtype-altfmt.d: Ditto.
* testsuite/gas/riscv/vector-vtype-altfmt.s: Ditto.
* testsuite/gas/riscv/zvdota.d: Ditto.
* testsuite/gas/riscv/zvdota.s: Ditto.
include/ChangeLog:
* opcode/riscv-opc.h (MATCH_VQWDOTAU_VV, MASK_VQWDOTAU_VV): New encoding.
(MATCH_VQWDOTAS_VV, MASK_VQWDOTAS_VV): Ditto.
(MATCH_VFWDOTA_VV, MASK_VFWDOTA_VV): Ditto.
(MATCH_VFQWDOTA_VV, MASK_VFQWDOTA_VV): Ditto.
(MATCH_VFQWDOTA_ALT_VV, MASK_VFQWDOTA_ALT_VV): Ditto.
(DECLARE_INSN): New declarations.
* opcode/riscv.h (OP_MASK_VTYPE_ALTFMT, OP_SH_VTYPE_ALTFMT): New vtype altfmt.
(enum riscv_insn_class): New extension classes.
opcodes/ChangeLog:
* riscv-dis.c (print_insn_args): Add altfmt support.
* riscv-opc.c (riscv_opcodes): New instructions.
{"zvfhmin", "+zve32f", check_implicit_always},
{"zvfbfwma", "+zve32f,+zfbfmin", check_implicit_always},
{"zvfbfmin", "+zve32f", check_implicit_always},
+ {"zvfqwdota8f", "+zve32f", check_implicit_always},
+ {"zvfwdota16bf", "+zve32f", check_implicit_always},
+ {"zvqwdota8i", "+zve32x", check_implicit_always},
+ {"zvqwdota16i", "+zve64x", check_implicit_always},
{"zve64d", "+d,+zve64f", check_implicit_always},
{"zve64f", "+zve32f,+zve64x,+zvl64b", check_implicit_always},
{"zve32f", "+f,+zve32x,+zvl32b", check_implicit_always},
{"zvfbfwma", ISA_SPEC_CLASS_DRAFT, 1, 0, 0 },
{"zvfh", ISA_SPEC_CLASS_DRAFT, 1, 0, 0 },
{"zvfhmin", ISA_SPEC_CLASS_DRAFT, 1, 0, 0 },
+ {"zvfqwdota8f", ISA_SPEC_CLASS_DRAFT, 1, 0, 0 },
+ {"zvfwdota16bf", ISA_SPEC_CLASS_DRAFT, 1, 0, 0 },
{"zvkb", ISA_SPEC_CLASS_DRAFT, 1, 0, 0 },
{"zvkg", ISA_SPEC_CLASS_DRAFT, 1, 0, 0 },
{"zvkn", ISA_SPEC_CLASS_DRAFT, 1, 0, 0 },
{"zvksg", ISA_SPEC_CLASS_DRAFT, 1, 0, 0 },
{"zvksc", ISA_SPEC_CLASS_DRAFT, 1, 0, 0 },
{"zvkt", ISA_SPEC_CLASS_DRAFT, 1, 0, 0 },
+ {"zvqwdota8i", ISA_SPEC_CLASS_DRAFT, 1, 0, 0 },
+ {"zvqwdota16i", ISA_SPEC_CLASS_DRAFT, 1, 0, 0 },
{"zvl32b", ISA_SPEC_CLASS_DRAFT, 1, 0, 0 },
{"zvl64b", ISA_SPEC_CLASS_DRAFT, 1, 0, 0 },
{"zvl128b", ISA_SPEC_CLASS_DRAFT, 1, 0, 0 },
return riscv_subset_supports (rps, "zvfbfmin");
case INSN_CLASS_ZVFBFWMA:
return riscv_subset_supports (rps, "zvfbfwma");
+ case INSN_CLASS_ZVFQWDOTA8F:
+ return riscv_subset_supports (rps, "zvfqwdota8f");
+ case INSN_CLASS_ZVFWDOTA16BF:
+ return riscv_subset_supports (rps, "zvfwdota16bf");
case INSN_CLASS_ZVKB:
return riscv_subset_supports (rps, "zvkb");
case INSN_CLASS_ZVKG:
return riscv_subset_supports (rps, "zvksed");
case INSN_CLASS_ZVKSH:
return riscv_subset_supports (rps, "zvksh");
+ case INSN_CLASS_ZVQWDOTA8I:
+ return riscv_subset_supports (rps, "zvqwdota8i");
+ case INSN_CLASS_ZVQWDOTA16I:
+ return riscv_subset_supports (rps, "zvqwdota16i");
+ case INSN_CLASS_ZVQWDOTA8I_OR_ZVQWDOTA16I:
+ return (riscv_subset_supports (rps, "zvqwdota8i")
+ || riscv_subset_supports (rps, "zvqwdota16i"));
case INSN_CLASS_ZCB:
return riscv_subset_supports (rps, "zcb");
case INSN_CLASS_ZCB_AND_ZBB:
return "zvfbfmin";
case INSN_CLASS_ZVFBFWMA:
return "zvfbfwma";
+ case INSN_CLASS_ZVFQWDOTA8F:
+ return "zvfqwdota8f";
+ case INSN_CLASS_ZVFWDOTA16BF:
+ return "zvfwdota16bf";
case INSN_CLASS_ZVKB:
return _("zvkb");
case INSN_CLASS_ZVKG:
return _("zvksed");
case INSN_CLASS_ZVKSH:
return _("zvksh");
+ case INSN_CLASS_ZVQWDOTA8I:
+ return "zvqwdota8i";
+ case INSN_CLASS_ZVQWDOTA16I:
+ return "zvqwdota16i";
+ case INSN_CLASS_ZVQWDOTA8I_OR_ZVQWDOTA16I:
+ return _("zvqwdota8i' or `zvqwdota16i");
case INSN_CLASS_ZCB:
return "zcb";
case INSN_CLASS_ZCB_AND_ZBA:
and will be removed in a future release.
* Add support for RISC-V standard extensions:
- zalasr v1.0, svrsw60t59b v1.0, zvabd v1.0, smpmpmt v1.0.
+ zalasr v1.0, svrsw60t59b v1.0, zvabd v1.0, smpmpmt v1.0,
+ zvqwdota8i, zvqwdota16i, zvfwdota16bf and zvfqwdota8f v1.0.
* Add support for RISC-V vendor extensions:
SpacemiT: xsmtvdot v1.0, xsmtvdotii v1.0.
On exit, EXPR_PARSE_END points to the first character after the
expression. */
+static bool
+riscv_vtype_altfmt_supported (void)
+{
+ return (riscv_subset_supports (&riscv_rps_as, "zvfbfa")
+ || riscv_subset_supports (&riscv_rps_as, "zvfofp8min")
+ || riscv_subset_supports (&riscv_rps_as, "zvfqwdota8f")
+ || riscv_subset_supports (&riscv_rps_as, "zvfwdota16bf")
+ || riscv_subset_supports (&riscv_rps_as, "zvqwdota8i")
+ || riscv_subset_supports (&riscv_rps_as, "zvqwdota16i"));
+}
+
static void
my_getVsetvliExpression (expressionS *ep, char *str)
{
unsigned int vsew_value = 0, vlmul_value = 0;
- unsigned int vta_value = 0, vma_value = 0;
+ unsigned int vta_value = 0, vma_value = 0, altfmt_value = 0;
bool vsew_found = false, vlmul_found = false;
bool vta_found = false, vma_found = false;
- if (arg_lookup (&str, riscv_vsew, ARRAY_SIZE (riscv_vsew), &vsew_value))
+ if (arg_lookup (&str, riscv_vsew_altfmt,
+ ARRAY_SIZE (riscv_vsew_altfmt), &vsew_value))
+ {
+ if (*str == ',')
+ ++str;
+ if (vsew_found)
+ as_bad (_("multiple vsew constants"));
+ if (!riscv_vtype_altfmt_supported ())
+ as_bad (_("symbolic vtype altfmt requires `zvfbfa', "
+ "`zvfofp8min' or a `zvdota' extension"));
+ altfmt_value = 1 << OP_SH_VTYPE_ALTFMT;
+ vsew_found = true;
+ }
+ if (!vsew_found
+ && arg_lookup (&str, riscv_vsew, ARRAY_SIZE (riscv_vsew), &vsew_value))
{
if (*str == ',')
++str;
ep->X_add_number = (vlmul_value << OP_SH_VLMUL)
| (vsew_value << OP_SH_VSEW)
| (vta_value << OP_SH_VTA)
- | (vma_value << OP_SH_VMA);
+ | (vma_value << OP_SH_VMA)
+ | altfmt_value;
expr_parse_end = str;
}
else
zvfbfwma 1.0
zvfh 1.0
zvfhmin 1.0
+ zvfqwdota8f 1.0
+ zvfwdota16bf 1.0
zvkb 1.0
zvkg 1.0
zvkn 1.0
zvksg 1.0
zvksc 1.0
zvkt 1.0
+ zvqwdota8i 1.0
+ zvqwdota16i 1.0
zvl32b 1.0
zvl64b 1.0
zvl128b 1.0
--- /dev/null
+#as: -march=rv64gc_zvfqwdota8f
+#objdump: -dr
+
+.*:[ ]+file format .*
+
+
+Disassembly of section .text:
+
+0+000 <target>:
+[ ]+[0-9a-f]+:[ ]+10707057[ ]+vsetvli[ ]+zero,zero,e8alt,mf2,tu,mu
+[ ]+[0-9a-f]+:[ ]+10707057[ ]+vsetvli[ ]+zero,zero,e8alt,mf2,tu,mu
+[ ]+[0-9a-f]+:[ ]+10707057[ ]+vsetvli[ ]+zero,zero,e8alt,mf2,tu,mu
+[ ]+[0-9a-f]+:[ ]+10f07057[ ]+vsetvli[ ]+zero,zero,e16alt,mf2,tu,mu
+[ ]+[0-9a-f]+:[ ]+10f07057[ ]+vsetvli[ ]+zero,zero,e16alt,mf2,tu,mu
+[ ]+[0-9a-f]+:[ ]+10f07057[ ]+vsetvli[ ]+zero,zero,e16alt,mf2,tu,mu
--- /dev/null
+target:
+ vsetvli zero, zero, e8alt, mf2, tu, mu
+ vsetvli zero, zero, e8alt, mf2, tu, mu
+ vsetvli zero, zero, e8alt, mf2, tu, mu
+ vsetvli zero, zero, e16alt, mf2, tu, mu
+ vsetvli zero, zero, e16alt, mf2, tu, mu
+ vsetvli zero, zero, e16alt, mf2, tu, mu
--- /dev/null
+#as: -march=rv64gc_zvqwdota8i_zvqwdota16i_zvfwdota16bf_zvfqwdota8f
+#objdump: -dr
+
+.*:[ ]+file format .*
+
+
+Disassembly of section .text:
+
+0+000 <target>:
+[ ]+[0-9a-f]+:[ ]+9a8f8877[ ]+vqwdotau.vv[ ]+v16,v8,v31
+[ ]+[0-9a-f]+:[ ]+9e8f8877[ ]+vqwdotas.vv[ ]+v16,v8,v31
+[ ]+[0-9a-f]+:[ ]+928f9877[ ]+vfwdota.vv[ ]+v16,v8,v31
+[ ]+[0-9a-f]+:[ ]+9a8f9877[ ]+vfqwdota.vv[ ]+v16,v8,v31
+[ ]+[0-9a-f]+:[ ]+9e8f9877[ ]+vfqwdota.alt.vv[ ]+v16,v8,v31
+[ ]+[0-9a-f]+:[ ]+988f8877[ ]+vqwdotau.vv[ ]+v16,v8,v31
+[ ]+[0-9a-f]+:[ ]+9c8f8877[ ]+vqwdotas.vv[ ]+v16,v8,v31
+[ ]+[0-9a-f]+:[ ]+928f9877[ ]+vfwdota.vv[ ]+v16,v8,v31
+[ ]+[0-9a-f]+:[ ]+980f9877[ ]+vfqwdota.vv[ ]+v16,v0,v31,v0.t
+[ ]+[0-9a-f]+:[ ]+9c0f9877[ ]+vfqwdota.alt.vv[ ]+v16,v0,v31,v0.t
--- /dev/null
+target:
+ vqwdotau.vv v16, v8, v31
+ vqwdotas.vv v16, v8, v31
+ vfwdota.vv v16, v8, v31
+ vfqwdota.vv v16, v8, v31
+ vfqwdota.alt.vv v16, v8, v31
+ vqwdotau.vv v16, v8, v31, v0.t
+ vqwdotas.vv v16, v8, v31, v0.t
+ vfwdota.vv v16, v8, v31, v0.t
+ vfqwdota.vv v16, v0, v31, v0.t
+ vfqwdota.alt.vv v16, v0, v31, v0.t
#define MASK_VFWMACCBF16_VF 0xfc00707f
#define MATCH_VFWMACCBF16_VV 0xec001057
#define MASK_VFWMACCBF16_VV 0xfc00707f
+/* Zvdota instructions. */
+#define MATCH_VQWDOTAU_VV 0x98000077
+#define MASK_VQWDOTAU_VV 0xfc00707f
+#define MATCH_VQWDOTAS_VV 0x9c000077
+#define MASK_VQWDOTAS_VV 0xfc00707f
+#define MATCH_VFWDOTA_VV 0x90001077
+#define MASK_VFWDOTA_VV 0xfc00707f
+#define MATCH_VFQWDOTA_VV 0x98001077
+#define MASK_VFQWDOTA_VV 0xfc00707f
+#define MATCH_VFQWDOTA_ALT_VV 0x9c001077
+#define MASK_VFQWDOTA_ALT_VV 0xfc00707f
/* Vendor-specific (CORE-V) Xcvmac instructions. */
#define MATCH_CV_MAC 0x9000302b
#define MASK_CV_MAC 0xfe00707f
/* Zvfbfwma instructions. */
DECLARE_INSN(VFWMACCBF16_VF, MATCH_VFWMACCBF16_VF, MASK_VFWMACCBF16_VF)
DECLARE_INSN(VFWMACCBF16_VV, MATCH_VFWMACCBF16_VV, MASK_VFWMACCBF16_VV)
+/* Zvdota instructions. */
+DECLARE_INSN(vqwdotau_vv, MATCH_VQWDOTAU_VV, MASK_VQWDOTAU_VV)
+DECLARE_INSN(vqwdotas_vv, MATCH_VQWDOTAS_VV, MASK_VQWDOTAS_VV)
+DECLARE_INSN(vfwdota_vv, MATCH_VFWDOTA_VV, MASK_VFWDOTA_VV)
+DECLARE_INSN(vfqwdota_vv, MATCH_VFQWDOTA_VV, MASK_VFQWDOTA_VV)
+DECLARE_INSN(vfqwdota_alt_vv, MATCH_VFQWDOTA_ALT_VV, MASK_VFQWDOTA_ALT_VV)
/* Zvbb/Zvkb instructions. */
DECLARE_INSN(vandn_vv, MATCH_VANDN_VV, MASK_VANDN_VV)
DECLARE_INSN(vandn_vx, MATCH_VANDN_VX, MASK_VANDN_VX)
#define OP_SH_VTA 6
#define OP_MASK_VMA 0x1
#define OP_SH_VMA 7
+#define OP_MASK_VTYPE_ALTFMT 0x1
+#define OP_SH_VTYPE_ALTFMT 8
#define OP_MASK_VWD 0x1
#define OP_SH_VWD 26
INSN_CLASS_ZVBC,
INSN_CLASS_ZVFBFMIN,
INSN_CLASS_ZVFBFWMA,
+ INSN_CLASS_ZVFQWDOTA8F,
+ INSN_CLASS_ZVFWDOTA16BF,
INSN_CLASS_ZVKB,
INSN_CLASS_ZVKG,
INSN_CLASS_ZVKNED,
INSN_CLASS_ZVKSED,
INSN_CLASS_ZVKSH,
INSN_CLASS_ZVABD,
+ INSN_CLASS_ZVQWDOTA8I,
+ INSN_CLASS_ZVQWDOTA16I,
+ INSN_CLASS_ZVQWDOTA8I_OR_ZVQWDOTA16I,
INSN_CLASS_ZICFISS,
INSN_CLASS_ZICFISS_AND_ZCMOP,
INSN_CLASS_ZICFILP,
extern const char riscv_vecr_names_numeric[NVECR][NRC];
extern const char riscv_vecm_names_numeric[NVECM][NRC];
extern const char * const riscv_vsew[8];
+extern const char * const riscv_vsew_altfmt[2];
extern const char * const riscv_vlmul[8];
extern const char * const riscv_vta[2];
extern const char * const riscv_vma[2];
unsigned int imm_vsew = EXTRACT_OPERAND (VSEW, imm);
unsigned int imm_vta = EXTRACT_OPERAND (VTA, imm);
unsigned int imm_vma = EXTRACT_OPERAND (VMA, imm);
- unsigned int imm_vtype_res = (imm >> 8);
+ unsigned int imm_vtype_altfmt
+ = EXTRACT_OPERAND (VTYPE_ALTFMT, imm);
+ unsigned int imm_vtype_res = (imm >> 9);
if (imm_vsew < ARRAY_SIZE (riscv_vsew)
&& imm_vlmul < ARRAY_SIZE (riscv_vlmul)
&& imm_vta < ARRAY_SIZE (riscv_vta)
&& imm_vma < ARRAY_SIZE (riscv_vma)
&& !imm_vtype_res
+ && (imm_vtype_altfmt == 0
+ || imm_vsew < ARRAY_SIZE (riscv_vsew_altfmt))
&& riscv_vsew[imm_vsew] != NULL
&& riscv_vlmul[imm_vlmul] != NULL)
print (info->stream, dis_style_text, "%s,%s,%s,%s",
- riscv_vsew[imm_vsew],
+ imm_vtype_altfmt
+ ? riscv_vsew_altfmt[imm_vsew]
+ : riscv_vsew[imm_vsew],
riscv_vlmul[imm_vlmul], riscv_vta[imm_vta],
riscv_vma[imm_vma]);
else
"e8", "e16", "e32", "e64", NULL, NULL, NULL, NULL
};
+/* The vsetvli/vsetivli vsew altfmt constants. */
+const char * const riscv_vsew_altfmt[2] =
+{
+ "e8alt", "e16alt"
+};
+
/* The vsetvli vlmul constants. */
const char * const riscv_vlmul[8] =
{
{"vfwmaccbf16.vf", 0, INSN_CLASS_ZVFBFWMA, "Vd,S,VtVm", MATCH_VFWMACCBF16_VF, MASK_VFWMACCBF16_VF, match_opcode, 0},
{"vfwmaccbf16.vv", 0, INSN_CLASS_ZVFBFWMA, "Vd,Vs,VtVm", MATCH_VFWMACCBF16_VV, MASK_VFWMACCBF16_VV, match_opcode, 0},
+/* Zvdota instructions. */
+{"vqwdotau.vv", 0, INSN_CLASS_ZVQWDOTA8I_OR_ZVQWDOTA16I, "Vd,Vt,VsVm", MATCH_VQWDOTAU_VV, MASK_VQWDOTAU_VV, match_opcode, 0},
+{"vqwdotas.vv", 0, INSN_CLASS_ZVQWDOTA8I_OR_ZVQWDOTA16I, "Vd,Vt,VsVm", MATCH_VQWDOTAS_VV, MASK_VQWDOTAS_VV, match_opcode, 0},
+{"vfwdota.vv", 0, INSN_CLASS_ZVFWDOTA16BF, "Vd,Vt,VsVm", MATCH_VFWDOTA_VV, MASK_VFWDOTA_VV, match_opcode, 0},
+{"vfqwdota.vv", 0, INSN_CLASS_ZVFQWDOTA8F, "Vd,Vt,VsVm", MATCH_VFQWDOTA_VV, MASK_VFQWDOTA_VV, match_opcode, 0},
+{"vfqwdota.alt.vv", 0, INSN_CLASS_ZVFQWDOTA8F, "Vd,Vt,VsVm", MATCH_VFQWDOTA_ALT_VV, MASK_VFQWDOTA_ALT_VV, match_opcode, 0},
+
/* Zvbb/Zvkb instructions. */
{"vandn.vv", 0, INSN_CLASS_ZVKB, "Vd,Vt,VsVm", MATCH_VANDN_VV, MASK_VANDN_VV, match_opcode, 0},
{"vandn.vx", 0, INSN_CLASS_ZVKB, "Vd,Vt,sVm", MATCH_VANDN_VX, MASK_VANDN_VX, match_opcode, 0},