1 files changed, 127 insertions, 0 deletions
diff --git a/vendor/github.com/klauspost/compress/zstd/fse_decoder_amd64.s b/vendor/github.com/klauspost/compress/zstd/fse_decoder_amd64.s
new file mode 100644
index 000000000..da32b4420
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/zstd/fse_decoder_amd64.s
@@ -0,0 +1,127 @@
+// Code generated by command: go run gen_fse.go -out ../fse_decoder_amd64.s -pkg=zstd. DO NOT EDIT.
+
+//go:build !appengine && !noasm && gc && !noasm
+// +build !appengine,!noasm,gc,!noasm
+
+// func buildDtable_asm(s *fseDecoder, ctx *buildDtableAsmContext) int
+TEXT ·buildDtable_asm(SB), $0-24
+	MOVQ ctx+8(FP), CX
+	MOVQ s+0(FP), DI
+
+	// Load values
+	MOVBQZX 4098(DI), DX
+	XORQ    AX, AX
+	BTSQ    DX, AX
+	MOVQ    (CX), BX
+	MOVQ    16(CX), SI
+	LEAQ    -1(AX), R8
+	MOVQ    8(CX), CX
+	MOVWQZX 4096(DI), DI
+
+	// End load values
+	// Init, lay down lowprob symbols
+	XORQ R9, R9
+	JMP  init_main_loop_condition
+
+init_main_loop:
+	MOVWQSX (CX)(R9*2), R10
+	CMPW    R10, $-1
+	JNE     do_not_update_high_threshold
+	MOVB    R9, 1(SI)(R8*8)
+	DECQ    R8
+	MOVQ    $0x0000000000000001, R10
+
+do_not_update_high_threshold:
+	MOVW R10, (BX)(R9*2)
+	INCQ R9
+
+init_main_loop_condition:
+	CMPQ R9, DI
+	JL   init_main_loop
+
+	// Spread symbols
+	// Calculate table step
+	MOVQ AX, R9
+	SHRQ $0x01, R9
+	MOVQ AX, R10
+	SHRQ $0x03, R10
+	LEAQ 3(R9)(R10*1), R9
+
+	// Fill add bits values
+	LEAQ -1(AX), R10
+	XORQ R11, R11
+	XORQ R12, R12
+	JMP  spread_main_loop_condition
+
+spread_main_loop:
+	XORQ    R13, R13
+	MOVWQSX (CX)(R12*2), R14
+	JMP     spread_inner_loop_condition
+
+spread_inner_loop:
+	MOVB R12, 1(SI)(R11*8)
+
+adjust_position:
+	ADDQ R9, R11
+	ANDQ R10, R11
+	CMPQ R11, R8
+	JG   adjust_position
+	INCQ R13
+
+spread_inner_loop_condition:
+	CMPQ R13, R14
+	JL   spread_inner_loop
+	INCQ R12
+
+spread_main_loop_condition:
+	CMPQ  R12, DI
+	JL    spread_main_loop
+	TESTQ R11, R11
+	JZ    spread_check_ok
+	MOVQ  ctx+8(FP), AX
+	MOVQ  R11, 24(AX)
+	MOVQ  $+1, ret+16(FP)
+	RET
+
+spread_check_ok:
+	// Build Decoding table
+	XORQ DI, DI
+
+build_table_main_table:
+	MOVBQZX 1(SI)(DI*8), CX
+	MOVWQZX (BX)(CX*2), R8
+	LEAQ    1(R8), R9
+	MOVW    R9, (BX)(CX*2)
+	MOVQ    R8, R9
+	BSRQ    R9, R9
+	MOVQ    DX, CX
+	SUBQ    R9, CX
+	SHLQ    CL, R8
+	SUBQ    AX, R8
+	MOVB    CL, (SI)(DI*8)
+	MOVW    R8, 2(SI)(DI*8)
+	CMPQ    R8, AX
+	JLE     build_table_check1_ok
+	MOVQ    ctx+8(FP), CX
+	MOVQ    R8, 24(CX)
+	MOVQ    AX, 32(CX)
+	MOVQ    $+2, ret+16(FP)
+	RET
+
+build_table_check1_ok:
+	TESTB CL, CL
+	JNZ   build_table_check2_ok
+	CMPW  R8, DI
+	JNE   build_table_check2_ok
+	MOVQ  ctx+8(FP), AX
+	MOVQ  R8, 24(AX)
+	MOVQ  DI, 32(AX)
+	MOVQ  $+3, ret+16(FP)
+	RET
+
+build_table_check2_ok:
+	INCQ DI
+	CMPQ DI, AX
+	JL   build_table_main_table
+	MOVQ $+0, ret+16(FP)
+	RET