📄 zgemm_kernel_4x2_sse3.s

📁 Optimized GotoBLAS libraries
💻 S
📖 第 1 页 / 共 3 页
字号:
	KERNEL5 (32  *  0)	KERNEL6 (32  *  0)	KERNEL7 (32  *  0)	KERNEL8 (32  *  0)	KERNEL9 (32  *  0)	KERNEL10(32  *  0)	KERNEL11(32  *  0)	KERNEL12(32  *  0)	KERNEL13(32  *  0)	KERNEL14(32  *  0)	KERNEL15(32  *  0)	KERNEL16(32  *  0)	addq   $64 * SIZE, AO	addq   $64 * SIZE, BO	decq   %rax	jne    .L12#endif	ALIGN_4.L15:#ifndef TRMMKERNEL	movq	K, %rax#else	movq	KKK, %rax#endif	movaps	ALPHA_R, %xmm14	movaps	ALPHA_I, %xmm15	andq	$7, %rax		# if (k & 1)	BRANCH	je .L18	ALIGN_4.L16:	mulps	%xmm8, %xmm9	addps	%xmm9, %xmm0	movshdup  0 * SIZE(BO), %xmm9	mulps	%xmm8, %xmm9	ADDSUB	%xmm9, %xmm1	movsldup  4 * SIZE(BO), %xmm9	mulps	%xmm8, %xmm9	addps	%xmm9, %xmm2	movshdup  4 * SIZE(BO), %xmm9	mulps	%xmm8, %xmm9	movaps	  4 * SIZE(AO), %xmm8	ADDSUB	%xmm9, %xmm3	movsldup  0 * SIZE(BO), %xmm9	mulps	%xmm8, %xmm9	addps	%xmm9, %xmm4	movshdup  0 * SIZE(BO), %xmm9	mulps	%xmm8, %xmm9	ADDSUB	%xmm9, %xmm5	movsldup  4 * SIZE(BO), %xmm9	mulps	%xmm8, %xmm9	addps	%xmm9, %xmm6	movshdup  4 * SIZE(BO), %xmm9	mulps	%xmm8, %xmm9	movaps	  8 * SIZE(AO), %xmm8	ADDSUB	%xmm9, %xmm7	movsldup  8 * SIZE(BO), %xmm9	addq	$8 * SIZE, AO	addq	$8 * SIZE, BO	decq	%rax	jg	.L16	ALIGN_4.L18:#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \    defined(NR) || defined(NC) || defined(TR) || defined(TC)	shufps	$0xb1, %xmm1, %xmm1	shufps	$0xb1, %xmm3, %xmm3	shufps	$0xb1, %xmm5, %xmm5	shufps	$0xb1, %xmm7, %xmm7	addsubps	%xmm1, %xmm0	addsubps	%xmm3, %xmm2	addsubps	%xmm5, %xmm4	addsubps	%xmm7, %xmm6	movaps	%xmm0, %xmm1	movaps	%xmm2, %xmm3	movaps	%xmm4, %xmm5	movaps	%xmm6, %xmm7	shufps	$0xb1, %xmm0, %xmm0	shufps	$0xb1, %xmm2, %xmm2	shufps	$0xb1, %xmm4, %xmm4	shufps	$0xb1, %xmm6, %xmm6#else	shufps	$0xb1, %xmm0, %xmm0	shufps	$0xb1, %xmm2, %xmm2	shufps	$0xb1, %xmm4, %xmm4	shufps	$0xb1, %xmm6, %xmm6	addsubps	%xmm0, %xmm1	addsubps	%xmm2, %xmm3	addsubps	%xmm4, %xmm5	addsubps	%xmm6, %xmm7	movaps	%xmm1, %xmm0	movaps	%xmm3, %xmm2	movaps	%xmm5, %xmm4	movaps	%xmm7, %xmm6	shufps	$0xb1, %xmm1, %xmm1	shufps	$0xb1, %xmm3, %xmm3	shufps	$0xb1, %xmm5, %xmm5	shufps	$0xb1, %xmm7, %xmm7#endif	mulps	%xmm14, %xmm1	mulps	%xmm15, %xmm0	mulps	%xmm14, %xmm3	mulps	%xmm15, %xmm2	mulps	%xmm14, %xmm5	mulps	%xmm15, %xmm4	mulps	%xmm14, %xmm7	mulps	%xmm15, %xmm6	addps	%xmm1, %xmm0	addps	%xmm3, %xmm2	addps	%xmm5, %xmm4	addps	%xmm7, %xmm6#if! defined(TRMMKERNEL) && !defined(BETAZERO)	shufps	$0xe4, %xmm8,  %xmm8	shufps	$0xe4, %xmm9,  %xmm9	shufps	$0xe4, %xmm10, %xmm10	shufps	$0xe4, %xmm11, %xmm11	movsd	0 * SIZE(CO1), %xmm8	movhps	2 * SIZE(CO1), %xmm8	movsd	4 * SIZE(CO1), %xmm10	movhps	6 * SIZE(CO1), %xmm10	movsd	0 * SIZE(CO2), %xmm9	movhps	2 * SIZE(CO2), %xmm9	movsd	4 * SIZE(CO2), %xmm11	movhps	6 * SIZE(CO2), %xmm11	addps	%xmm8,  %xmm0	addps	%xmm9,  %xmm2	addps	%xmm10, %xmm4	addps	%xmm11, %xmm6#endif	movsd	%xmm0, 0 * SIZE(CO1)	movhps	%xmm0, 2 * SIZE(CO1)	movsd	%xmm4, 4 * SIZE(CO1)	movhps	%xmm4, 6 * SIZE(CO1)	movsd	%xmm2, 0 * SIZE(CO2)	movhps	%xmm2, 2 * SIZE(CO2)	movsd	%xmm6, 4 * SIZE(CO2)	movhps	%xmm6, 6 * SIZE(CO2)#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	movq	K, %rax	subq	KKK, %rax	leaq	(,%rax, 8), %rax	leaq	(AO, %rax, 4), AO	leaq	(BO, %rax, 4), BO#endif#if defined(TRMMKERNEL) && defined(LEFT)	addq	$4, KK#endif	addq	$8 * SIZE, CO1		# coffset += 4	addq	$8 * SIZE, CO2		# coffset += 4	decq	I			# i --	jg	.L11	ALIGN_4	.L20:	testq	$2, M	je	.L30#if !defined(TRMMKERNEL) || \	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	leaq	BUFFER, BO#else	leaq	BUFFER, BO	movq	KK, %rax	leaq	(, %rax,   8), %rax	leaq	(AO, %rax, 2), AO	leaq	(BO, %rax, 4), BO#endif		movaps	  0 * SIZE(AO), %xmm8	pxor	%xmm0, %xmm0	movaps	 16 * SIZE(AO), %xmm10	pxor	%xmm1, %xmm1	movsldup  0 * SIZE(BO), %xmm9	pxor	%xmm2, %xmm2	movsldup 16 * SIZE(BO), %xmm11	pxor	%xmm3, %xmm3	movsldup 32 * SIZE(BO), %xmm13	movsldup 48 * SIZE(BO), %xmm15#ifndef TRMMKERNEL	movq	K, %rax#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))	movq	K, %rax	subq	KK, %rax	movq	%rax, KKK	#else	movq	KK, %rax#ifdef LEFT	addq	$2, %rax#else	addq	$2, %rax#endif	movq	%rax, KKK#endif	sarq	$3, %rax	je	.L25	ALIGN_4.L22:	mulps	%xmm8, %xmm9	PREFETCH  (PREFETCHSIZE +  0) * SIZE(AO)	addps	%xmm9, %xmm0	movshdup  0 * SIZE(BO), %xmm9	mulps	%xmm8, %xmm9	ADDSUB	%xmm9, %xmm1	movsldup  4 * SIZE(BO), %xmm9	mulps	%xmm8, %xmm9	addps	%xmm9, %xmm2	movshdup  4 * SIZE(BO), %xmm9	mulps	%xmm8, %xmm9	movaps	  4 * SIZE(AO), %xmm8	ADDSUB	%xmm9, %xmm3	movsldup  8 * SIZE(BO), %xmm9	mulps	%xmm8, %xmm9	addps	%xmm9, %xmm0	movshdup  8 * SIZE(BO), %xmm9	mulps	%xmm8, %xmm9	ADDSUB	%xmm9, %xmm1	movsldup 12 * SIZE(BO), %xmm9	mulps	%xmm8, %xmm9	addps	%xmm9, %xmm2	movshdup 12 * SIZE(BO), %xmm9	mulps	%xmm8, %xmm9	movaps	  8 * SIZE(AO), %xmm8	ADDSUB	%xmm9, %xmm3	movsldup 64 * SIZE(BO), %xmm9	mulps	%xmm8, %xmm11	addps	%xmm11, %xmm0	movshdup 16 * SIZE(BO), %xmm11	mulps	%xmm8, %xmm11	ADDSUB	%xmm11, %xmm1	movsldup 20 * SIZE(BO), %xmm11	mulps	%xmm8, %xmm11	addps	%xmm11, %xmm2	movshdup 20 * SIZE(BO), %xmm11	mulps	%xmm8, %xmm11	movaps	 12 * SIZE(AO), %xmm8	ADDSUB	%xmm11, %xmm3	movsldup 24 * SIZE(BO), %xmm11	mulps	%xmm8, %xmm11	addps	%xmm11, %xmm0	movshdup 24 * SIZE(BO), %xmm11	mulps	%xmm8, %xmm11	ADDSUB	%xmm11, %xmm1	movsldup 28 * SIZE(BO), %xmm11	mulps	%xmm8, %xmm11	addps	%xmm11, %xmm2	movshdup 28 * SIZE(BO), %xmm11	mulps	%xmm8, %xmm11	movaps	 32 * SIZE(AO), %xmm8	ADDSUB	%xmm11, %xmm3	movsldup 80 * SIZE(BO), %xmm11	mulps	%xmm10, %xmm13	addps	%xmm13, %xmm0	movshdup 32 * SIZE(BO), %xmm13	mulps	%xmm10, %xmm13	ADDSUB	%xmm13, %xmm1	movsldup 36 * SIZE(BO), %xmm13	mulps	%xmm10, %xmm13	addps	%xmm13, %xmm2	movshdup 36 * SIZE(BO), %xmm13	mulps	%xmm10, %xmm13	movaps	 20 * SIZE(AO), %xmm10	ADDSUB	%xmm13, %xmm3	movsldup 40 * SIZE(BO), %xmm13	mulps	%xmm10, %xmm13	addps	%xmm13, %xmm0	movshdup 40 * SIZE(BO), %xmm13	mulps	%xmm10, %xmm13	ADDSUB	%xmm13, %xmm1	movsldup 44 * SIZE(BO), %xmm13	mulps	%xmm10, %xmm13	addps	%xmm13, %xmm2	movshdup 44 * SIZE(BO), %xmm13	mulps	%xmm10, %xmm13	movaps	 24 * SIZE(AO), %xmm10	ADDSUB	%xmm13, %xmm3	movsldup 96 * SIZE(BO), %xmm13	mulps	%xmm10, %xmm15	addps	%xmm15, %xmm0	movshdup 48 * SIZE(BO), %xmm15	mulps	%xmm10, %xmm15	ADDSUB	%xmm15, %xmm1	movsldup 52 * SIZE(BO), %xmm15	mulps	%xmm10, %xmm15	addps	%xmm15, %xmm2	movshdup 52 * SIZE(BO), %xmm15	mulps	%xmm10, %xmm15	movaps	 28 * SIZE(AO), %xmm10	ADDSUB	%xmm15, %xmm3	movsldup 56 * SIZE(BO), %xmm15	mulps	%xmm10, %xmm15	addps	%xmm15, %xmm0	movshdup 56 * SIZE(BO), %xmm15	mulps	%xmm10, %xmm15	ADDSUB	%xmm15, %xmm1	movsldup 60 * SIZE(BO), %xmm15	mulps	%xmm10, %xmm15	addps	%xmm15, %xmm2	movshdup 60 * SIZE(BO), %xmm15	mulps	%xmm10, %xmm15	movaps	 48 * SIZE(AO), %xmm10	ADDSUB	%xmm15, %xmm3	movsldup 112 * SIZE(BO), %xmm15	addq   $32 * SIZE, AO	addq   $64 * SIZE, BO	decq   %rax	jne    .L22	ALIGN_4.L25:#ifndef TRMMKERNEL	movq	K, %rax#else	movq	KKK, %rax#endif	movaps	ALPHA_R, %xmm14	movaps	ALPHA_I, %xmm15	andq	$7, %rax		# if (k & 1)	BRANCH	je .L28	ALIGN_4.L26:	mulps	%xmm8, %xmm9	addps	%xmm9, %xmm0	movshdup  0 * SIZE(BO), %xmm9	mulps	%xmm8, %xmm9	ADDSUB	%xmm9, %xmm1	movsldup  4 * SIZE(BO), %xmm9	mulps	%xmm8, %xmm9	addps	%xmm9, %xmm2	movshdup  4 * SIZE(BO), %xmm9	mulps	%xmm8, %xmm9	movaps	  4 * SIZE(AO), %xmm8	ADDSUB	%xmm9, %xmm3	movsldup  8 * SIZE(BO), %xmm9	addq	$ 4 * SIZE, AO	addq	$ 8 * SIZE, BO	decq	%rax	jg	.L26	ALIGN_4.L28:#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \    defined(NR) || defined(NC) || defined(TR) || defined(TC)	shufps	$0xb1, %xmm1, %xmm1	shufps	$0xb1, %xmm3, %xmm3	addsubps	%xmm1, %xmm0	addsubps	%xmm3, %xmm2	movaps	%xmm0, %xmm1	movaps	%xmm2, %xmm3	shufps	$0xb1, %xmm0, %xmm0	shufps	$0xb1, %xmm2, %xmm2#else	shufps	$0xb1, %xmm0, %xmm0	shufps	$0xb1, %xmm2, %xmm2	addsubps	%xmm0, %xmm1	addsubps	%xmm2, %xmm3	movaps	%xmm1, %xmm0	movaps	%xmm3, %xmm2	shufps	$0xb1, %xmm1, %xmm1	shufps	$0xb1, %xmm3, %xmm3#endif	mulps	%xmm14, %xmm1	mulps	%xmm15, %xmm0	mulps	%xmm14, %xmm3	mulps	%xmm15, %xmm2	addps	%xmm1, %xmm0	addps	%xmm3, %xmm2#if! defined(TRMMKERNEL) && !defined(BETAZERO)	shufps	$0xe4, %xmm8,  %xmm8	shufps	$0xe4, %xmm10, %xmm10	movsd	0 * SIZE(CO1), %xmm8	movhps	2 * SIZE(CO1), %xmm8	movsd	0 * SIZE(CO2), %xmm10	movhps	2 * SIZE(CO2), %xmm10	addps	%xmm8,  %xmm0	addps	%xmm10, %xmm2#endif	movsd	%xmm0, 0 * SIZE(CO1)	movhps	%xmm0, 2 * SIZE(CO1)	movsd	%xmm2, 0 * SIZE(CO2)	movhps	%xmm2, 2 * SIZE(CO2)#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	movq	K, %rax	subq	KKK, %rax	leaq	(,%rax,    8), %rax	leaq	(AO, %rax, 2), AO	leaq	(BO, %rax, 4), BO#endif#if defined(TRMMKERNEL) && defined(LEFT)	addq	$2, KK#endif	addq	$4 * SIZE, CO1		# coffset += 4	addq	$4 * SIZE, CO2		# coffset += 4	ALIGN_4		.L30:	testq	$1, M	je	.L39#if !defined(TRMMKERNEL) || \	(defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \	(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	leaq	BUFFER, BO#else	leaq	BUFFER, BO	movq	KK, %rax	leaq	(, %rax,   8), %rax	leaq	(AO, %rax, 1), AO	leaq	(BO, %rax, 4), BO#endif	 	movddup	  0 * SIZE(AO), %xmm8	pxor	%xmm0, %xmm0 	movddup	  8 * SIZE(AO), %xmm10	pxor	%xmm1, %xmm1	movsd     0 * SIZE(BO), %xmm9	pxor	%xmm2, %xmm2	movsd    16 * SIZE(BO), %xmm11	pxor	%xmm3, %xmm3	movsd    32 * SIZE(BO), %xmm13	movsd    48 * SIZE(BO), %xmm15#ifndef TRMMKERNEL	movq	K, %rax#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))	movq	K, %rax	subq	KK, %rax	movq	%rax, KKK	#else	movq	KK, %rax#ifdef LEFT	addq	$1, %rax#else	addq	$2, %rax#endif	movq	%rax, KKK#endif	sarq	$3, %rax	je	.L35	ALIGN_4.L32:	shufps	$0x50, %xmm9, %xmm9	PREFETCH  (PREFETCHSIZE +  0) * SIZE(AO)	mulps	%xmm8, %xmm9	addps	%xmm9, %xmm0	movsd     4 * SIZE(BO), %xmm9	shufps	$0x50, %xmm9, %xmm9	mulps	%xmm8, %xmm9 	movddup	  2 * SIZE(AO), %xmm8	addps	%xmm9, %xmm1	movsd     8 * SIZE(BO), %xmm9	shufps	$0x50, %xmm9, %xmm9	mulps	%xmm8, %xmm9	addps	%xmm9, %xmm0	movsd    12 * SIZE(BO), %xmm9	shufps	$0x50, %xmm9, %xmm9	mulps	%xmm8, %xmm9 	movddup	  4 * SIZE(AO), %xmm8	addps	%xmm9, %xmm1	movsd    64 * SIZE(BO), %xmm9	shufps	$0x50, %xmm11, %xmm11	mulps	%xmm8, %xmm11	addps	%xmm11, %xmm0	movsd    20 * SIZE(BO), %xmm11	shufps	$0x50, %xmm11, %xmm11	mulps	%xmm8, %xmm11 	movddup	  6 * SIZE(AO), %xmm8	addps	%xmm11, %xmm1	movsd    24 * SIZE(BO), %xmm11	shufps	$0x50, %xmm11, %xmm11	mulps	%xmm8, %xmm11	addps	%xmm11, %xmm0	movsd    28 * SIZE(BO), %xmm11	shufps	$0x50, %xmm11, %xmm11	mulps	%xmm8, %xmm11 	movddup	 16 * SIZE(AO), %xmm8	addps	%xmm11, %xmm1	movsd    80 * SIZE(BO), %xmm11	shufps	$0x50, %xmm13, %xmm13	mulps	%xmm10, %xmm13	addps	%xmm13, %xmm0	movsd    36 * SIZE(BO), %xmm13	shufps	$0x50, %xmm13, %xmm13	mulps	%xmm10, %xmm13 	movddup	 10 * SIZE(AO), %xmm10	addps	%xmm13, %xmm1	movsd    40 * SIZE(BO), %xmm13	shufps	$0x50, %xmm13, %xmm13	mulps	%xmm10, %xmm13	addps	%xmm13, %xmm0	movsd    44 * SIZE(BO), %xmm13	shufps	$0x50, %xmm13, %xmm13	mulps	%xmm10, %xmm13 	movddup	 12 * SIZE(AO), %xmm10	addps	%xmm13, %xmm1	movsd    96 * SIZE(BO), %xmm13	shufps	$0x50, %xmm15, %xmm15	mulps	%xmm10, %xmm15	addps	%xmm15, %xmm0	movsd    52 * SIZE(BO), %xmm15	shufps	$0x50, %xmm15, %xmm15	mulps	%xmm10, %xmm15 	movddup	 14 * SIZE(AO), %xmm10	addps	%xmm15, %xmm1	movsd    56 * SIZE(BO), %xmm15	shufps	$0x50, %xmm15, %xmm15	mulps	%xmm10, %xmm15	addps	%xmm15, %xmm0	movsd    60 * SIZE(BO), %xmm15	shufps	$0x50, %xmm15, %xmm15	mulps	%xmm10, %xmm15  	movddup	 24 * SIZE(AO), %xmm10	addps	%xmm15, %xmm1	movsd   112 * SIZE(BO), %xmm15	addq   $16 * SIZE, AO	addq   $64 * SIZE, BO	decq   %rax	jne    .L32	ALIGN_4.L35:#ifndef TRMMKERNEL	movq	K, %rax#else	movq	KKK, %rax#endif	movaps	ALPHA_R, %xmm14	movaps	ALPHA_I, %xmm15	andq	$7, %rax		# if (k & 1)	BRANCH	je .L38	ALIGN_4.L36:	shufps	$0x50, %xmm9, %xmm9	mulps	%xmm8, %xmm9	addps	%xmm9, %xmm0	movsd     4 * SIZE(BO), %xmm9	shufps	$0x50, %xmm9, %xmm9	mulps	%xmm8, %xmm9 	movddup	  2 * SIZE(AO), %xmm8	addps	%xmm9, %xmm1	movsd     8 * SIZE(BO), %xmm9	addq	$2 * SIZE, AO	addq	$8 * SIZE, BO	decq	%rax	jg	.L36	ALIGN_4.L38:	movaps	%xmm0, %xmm6	movlhps	%xmm1, %xmm0	movhlps %xmm6, %xmm1#if  defined(NR) || defined(NC) || defined(TR) || defined(TC) || \     defined(RR) || defined(RC) || defined(CR) || defined(CC) 	cmpeqps	%xmm7, %xmm7	pslld	$31,   %xmm7	xorps	%xmm7, %xmm1#endif          #if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \    defined(NR) || defined(NC) || defined(TR) || defined(TC)	shufps	$0xb1, %xmm1, %xmm1	addsubps	%xmm1, %xmm0	movaps	%xmm0, %xmm1	shufps	$0xb1, %xmm0, %xmm0#else	shufps	$0xb1, %xmm0, %xmm0	addsubps	%xmm0, %xmm1	movaps	%xmm1, %xmm0	shufps	$0xb1, %xmm1, %xmm1#endif	mulps	%xmm14, %xmm1	mulps	%xmm15, %xmm0	addps	%xmm1, %xmm0#if! defined(TRMMKERNEL) && !defined(BETAZERO)	movsd	0 * SIZE(CO1), %xmm8	movhps	0 * SIZE(CO2), %xmm8	addps	%xmm8, %xmm0#endif	movsd	%xmm0, 0 * SIZE(CO1)	movhps	%xmm0, 0 * SIZE(CO2)#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))	movq	K, %rax	subq	KKK, %rax	leaq	(,%rax,    8), %rax	leaq	(AO, %rax, 1), AO	leaq	(BO, %rax, 4), BO#endif#if defined(TRMMKERNEL) && defined(LEFT)	addq	$1, KK#endif	ALIGN_4	.L39:#if defined(TRMMKERNEL) && !defined(LEFT)	addl	$2, KK#endif	leaq	(C, LDC, 2), C		# c += 2 * ldc	decq	J			# j --	jg	.L01	ALIGN_4.L40:	testq	$1, N	je	.L999	ALIGN_4.L41:#if defined(TRMMKERNEL) && defined(LEFT)	movq	OFFSET, %rax	movq	%rax, KK#endif	/* Copying to Sub Buffer */	leaq	BUFFER, BO		movq	K, %rax	sarq	$3, %rax	jle	.L43	ALIGN_4	.L42:	movddup	 0 * SIZE(B), %xmm0	movddup	 2 * SIZE(B), %xmm1
⌨️ 快捷键说明

复制代码 Ctrl + C
搜索代码 Ctrl + F
全屏模式 F11
切换主题 Ctrl + Shift + D
显示快捷键 ?
增大字号 Ctrl + =
减小字号 Ctrl + -