FADD / FMUL / FFMA with modifiers (ptxas 12.9 for sm_90a, register
forms; Proof.FloatArithmeticContract holds both encoders to its words):
the opcode 0x221 / 0x220 / 0x223 (FMUL with bit 86 set, as plain FMUL),
the addend at bit 64 (FFMA), |right| bit 62, -right bit 63, -left bit 72,
|left| bit 73, -addend bit 75, .SAT bit 77, the rounding bits 78..79 (RN
0, RM 1, RP 2, RZ 3), .FTZ bit 80
979def sm86FloatOperationOpcode = (lambda unrestricted operation : (family SM86FloatOperation) .
980 (eliminate SM86FloatOperation (lambda unrestricted current : (family SM86FloatOperation) . Nat) operation
981 (branch SM86FloatOperationAdd . sm86OpcodeFloatAdd) (branch SM86FloatOperationMultiply . sm86OpcodeFloatMultiply)
982 (branch SM86FloatOperationFusedMultiplyAdd . sm86OpcodeFloatFusedMultiplyAdd)))The compiler supplied declaration spans and resolved links from this source snapshot. This page does not assert that this file belongs to a checked closure.