qemu-e2k/tests/tcg/hexagon/hvx_histogram_row.S
Taylor Simpson 49278c1b0d Hexagon HVX (tests/tcg/hexagon) histogram test
Acked-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Taylor Simpson <tsimpson@quicinc.com>
2021-11-03 16:01:38 -05:00

295 lines
7.2 KiB
ArmAsm

/*
* Copyright(c) 2021 Qualcomm Innovation Center, Inc. All Rights Reserved.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, see <http://www.gnu.org/licenses/>.
*/
/*
* void hvx_histogram_row(uint8_t *src, => r0
* int stride, => r1
* int width, => r2
* int height, => r3
* int *hist => r4)
*/
.text
.p2align 2
.global hvx_histogram_row
.type hvx_histogram_row, @function
hvx_histogram_row:
{ r2 = lsr(r2, #7) /* size / VLEN */
r5 = and(r2, #127) /* size % VLEN */
v1 = #0
v0 = #0
}
/*
* Step 1: Clean the whole vector register file
*/
{ v3:2 = v1:0
v5:4 = v1:0
p0 = cmp.gt(r2, #0) /* P0 = (width / VLEN > 0) */
p1 = cmp.eq(r5, #0) /* P1 = (width % VLEN == 0) */
}
{ q0 = vsetq(r5)
v7:6 = v1:0
}
{ v9:8 = v1:0
v11:10 = v1:0
}
{ v13:12 = v1:0
v15:14 = v1:0
}
{ v17:16 = v1:0
v19:18 = v1:0
}
{ v21:20 = v1:0
v23:22 = v1:0
}
{ v25:24 = v1:0
v27:26 = v1:0
}
{ v29:28 = v1:0
v31:30 = v1:0
r10 = add(r0, r1) /* R10 = &src[2 * stride] */
loop1(.outerloop, r3)
}
/*
* Step 2: vhist
*/
.falign
.outerloop:
{ if (!p0) jump .loopend
loop0(.innerloop, r2)
}
.falign
.innerloop:
{ v12.tmp = vmem(R0++#1)
vhist
}:endloop0
.falign
.loopend:
if (p1) jump .skip /* if (width % VLEN == 0) done with current row */
{ v13.tmp = vmem(r0 + #0)
vhist(q0)
}
.falign
.skip:
{ r0 = r10 /* R0 = &src[(i + 1) * stride] */
r10 = add(r10, r1) /* R10 = &src[(i + 2) * stride] */
}:endloop1
/*
* Step 3: Sum up the data
*/
{ v0.h = vshuff(v0.h)
r10 = ##0x00010001
}
v1.h = vshuff(v1.h)
{ V2.h = vshuff(v2.h)
v0.w = vdmpy(v0.h, r10.h):sat
}
{ v3.h = vshuff(v3.h)
v1.w = vdmpy(v1.h, r10.h):sat
}
{ v4.h = vshuff(V4.h)
v2.w = vdmpy(v2.h, r10.h):sat
}
{ v5.h = vshuff(v5.h)
v3.w = vdmpy(v3.h, r10.h):sat
}
{ v6.h = vshuff(v6.h)
v4.w = vdmpy(v4.h, r10.h):sat
}
{ v7.h = vshuff(v7.h)
v5.w = vdmpy(v5.h, r10.h):sat
}
{ v8.h = vshuff(V8.h)
v6.w = vdmpy(v6.h, r10.h):sat
}
{ v9.h = vshuff(V9.h)
v7.w = vdmpy(v7.h, r10.h):sat
}
{ v10.h = vshuff(v10.h)
v8.w = vdmpy(v8.h, r10.h):sat
}
{ v11.h = vshuff(v11.h)
v9.w = vdmpy(v9.h, r10.h):sat
}
{ v12.h = vshuff(v12.h)
v10.w = vdmpy(v10.h, r10.h):sat
}
{ v13.h = vshuff(V13.h)
v11.w = vdmpy(v11.h, r10.h):sat
}
{ v14.h = vshuff(v14.h)
v12.w = vdmpy(v12.h, r10.h):sat
}
{ v15.h = vshuff(v15.h)
v13.w = vdmpy(v13.h, r10.h):sat
}
{ v16.h = vshuff(v16.h)
v14.w = vdmpy(v14.h, r10.h):sat
}
{ v17.h = vshuff(v17.h)
v15.w = vdmpy(v15.h, r10.h):sat
}
{ v18.h = vshuff(v18.h)
v16.w = vdmpy(v16.h, r10.h):sat
}
{ v19.h = vshuff(v19.h)
v17.w = vdmpy(v17.h, r10.h):sat
}
{ v20.h = vshuff(v20.h)
v18.W = vdmpy(v18.h, r10.h):sat
}
{ v21.h = vshuff(v21.h)
v19.w = vdmpy(v19.h, r10.h):sat
}
{ v22.h = vshuff(v22.h)
v20.w = vdmpy(v20.h, r10.h):sat
}
{ v23.h = vshuff(v23.h)
v21.w = vdmpy(v21.h, r10.h):sat
}
{ v24.h = vshuff(v24.h)
v22.w = vdmpy(v22.h, r10.h):sat
}
{ v25.h = vshuff(v25.h)
v23.w = vdmpy(v23.h, r10.h):sat
}
{ v26.h = vshuff(v26.h)
v24.w = vdmpy(v24.h, r10.h):sat
}
{ v27.h = vshuff(V27.h)
v25.w = vdmpy(v25.h, r10.h):sat
}
{ v28.h = vshuff(v28.h)
v26.w = vdmpy(v26.h, r10.h):sat
}
{ v29.h = vshuff(v29.h)
v27.w = vdmpy(v27.h, r10.h):sat
}
{ v30.h = vshuff(v30.h)
v28.w = vdmpy(v28.h, r10.h):sat
}
{ v31.h = vshuff(v31.h)
v29.w = vdmpy(v29.h, r10.h):sat
r28 = #32
}
{ vshuff(v1, v0, r28)
v30.w = vdmpy(v30.h, r10.h):sat
}
{ vshuff(v3, v2, r28)
v31.w = vdmpy(v31.h, r10.h):sat
}
{ vshuff(v5, v4, r28)
v0.w = vadd(v1.w, v0.w)
v2.w = vadd(v3.w, v2.w)
}
{ vshuff(v7, v6, r28)
r7 = #64
}
{ vshuff(v9, v8, r28)
v4.w = vadd(v5.w, v4.w)
v6.w = vadd(v7.w, v6.w)
}
vshuff(v11, v10, r28)
{ vshuff(v13, v12, r28)
v8.w = vadd(v9.w, v8.w)
v10.w = vadd(v11.w, v10.w)
}
vshuff(v15, v14, r28)
{ vshuff(v17, v16, r28)
v12.w = vadd(v13.w, v12.w)
v14.w = vadd(v15.w, v14.w)
}
vshuff(v19, v18, r28)
{ vshuff(v21, v20, r28)
v16.w = vadd(v17.w, v16.w)
v18.w = vadd(v19.w, v18.w)
}
vshuff(v23, v22, r28)
{ vshuff(v25, v24, r28)
v20.w = vadd(v21.w, v20.w)
v22.w = vadd(v23.w, v22.w)
}
vshuff(v27, v26, r28)
{ vshuff(v29, v28, r28)
v24.w = vadd(v25.w, v24.w)
v26.w = vadd(v27.w, v26.w)
}
vshuff(v31, v30, r28)
{ v28.w = vadd(v29.w, v28.w)
vshuff(v2, v0, r7)
}
{ v30.w = vadd(v31.w, v30.w)
vshuff(v6, v4, r7)
v0.w = vadd(v0.w, v2.w)
}
{ vshuff(v10, v8, r7)
v1.tmp = vmem(r4 + #0) /* update hist[0-31] */
v0.w = vadd(v0.w, v1.w)
vmem(r4++#1) = v0.new
}
{ vshuff(v14, v12, r7)
v4.w = vadd(v4.w, v6.w)
v8.w = vadd(v8.w, v10.w)
}
{ vshuff(v18, v16, r7)
v1.tmp = vmem(r4 + #0) /* update hist[32-63] */
v4.w = vadd(v4.w, v1.w)
vmem(r4++#1) = v4.new
}
{ vshuff(v22, v20, r7)
v12.w = vadd(v12.w, v14.w)
V16.w = vadd(v16.w, v18.w)
}
{ vshuff(v26, v24, r7)
v1.tmp = vmem(r4 + #0) /* update hist[64-95] */
v8.w = vadd(v8.w, v1.w)
vmem(r4++#1) = v8.new
}
{ vshuff(v30, v28, r7)
v1.tmp = vmem(r4 + #0) /* update hist[96-127] */
v12.w = vadd(v12.w, v1.w)
vmem(r4++#1) = v12.new
}
{ v20.w = vadd(v20.w, v22.w)
v1.tmp = vmem(r4 + #0) /* update hist[128-159] */
v16.w = vadd(v16.w, v1.w)
vmem(r4++#1) = v16.new
}
{ v24.w = vadd(v24.w, v26.w)
v1.tmp = vmem(r4 + #0) /* update hist[160-191] */
v20.w = vadd(v20.w, v1.w)
vmem(r4++#1) = v20.new
}
{ v28.w = vadd(v28.w, v30.w)
v1.tmp = vmem(r4 + #0) /* update hist[192-223] */
v24.w = vadd(v24.w, v1.w)
vmem(r4++#1) = v24.new
}
{ v1.tmp = vmem(r4 + #0) /* update hist[224-255] */
v28.w = vadd(v28.w, v1.w)
vmem(r4++#1) = v28.new
}
jumpr r31
.size hvx_histogram_row, .-hvx_histogram_row