dilim/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2026-08-09 17:39:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

avcodec/fmtconvert: Remove int32_to_float_fmul_array8

Browse Source 
Unused since aebf07075f.

Reviewed-by: Martin Storsjö <martin@martin.st>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
This commit is contained in:
Andreas Rheinhardt
2026-08-04 20:47:22 +02:00
parent fc02470c62
commit 95c43d7df7
12 changed files with 0 additions and 355 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
libavcodec/aarch64/fmtconvert_init.c
View File

@@ -24,9 +24,6 @@
#include "libavutil/aarch64/cpu.h"
#include "libavcodec/fmtconvert.h"
void ff_int32_to_float_fmul_array8_neon(FmtConvertContext *c, float *dst,
const int32_t *src, const float *mul,
int len);
void ff_int32_to_float_fmul_scalar_neon(float *dst, const int32_t *src,
float mul, int len);
@@ -35,7 +32,6 @@ av_cold void ff_fmt_convert_init_aarch64(FmtConvertContext *c)
int cpu_flags = av_get_cpu_flags();
if (have_neon(cpu_flags)) {
c->int32_to_float_fmul_array8 = ff_int32_to_float_fmul_array8_neon;
c->int32_to_float_fmul_scalar = ff_int32_to_float_fmul_scalar_neon;
}
}

33
libavcodec/aarch64/fmtconvert_neon.S
View File

@@ -41,36 +41,3 @@ function ff_int32_to_float_fmul_scalar_neon, export=1
st1 {v3.4s,v4.4s}, [x0]
ret
endfunc
function ff_int32_to_float_fmul_array8_neon, export=1
lsr w4, w4, #3
subs w5, w4, #1
b.eq 1f
2:
ld1 {v0.4s,v1.4s}, [x2], #32
ld1 {v2.4s,v3.4s}, [x2], #32
scvtf v0.4s, v0.4s
scvtf v1.4s, v1.4s
ld1 {v16.2s}, [x3], #8
scvtf v2.4s, v2.4s
scvtf v3.4s, v3.4s
fmul v4.4s, v0.4s, v16.s[0]
fmul v5.4s, v1.4s, v16.s[0]
fmul v6.4s, v2.4s, v16.s[1]
fmul v7.4s, v3.4s, v16.s[1]
st1 {v4.4s,v5.4s}, [x1], #32
st1 {v6.4s,v7.4s}, [x1], #32
subs w5, w5, #2
b.gt 2b
b.eq 1f
ret
1:
ld1 {v0.4s,v1.4s}, [x2]
ld1 {v16.s}[0], [x3]
scvtf v0.4s, v0.4s
scvtf v1.4s, v1.4s
fmul v4.4s, v0.4s, v16.s[0]
fmul v5.4s, v1.4s, v16.s[0]
st1 {v4.4s,v5.4s}, [x1]
ret
endfunc

8
libavcodec/arm/fmtconvert_init_arm.c
View File

@@ -24,17 +24,11 @@
#include "libavutil/arm/cpu.h"
#include "libavcodec/fmtconvert.h"
void ff_int32_to_float_fmul_array8_neon(FmtConvertContext *c, float *dst,
const int32_t *src, const float *mul,
int len);
void ff_int32_to_float_fmul_scalar_neon(float *dst, const int32_t *src,
float mul, int len);
void ff_int32_to_float_fmul_scalar_vfp(float *dst, const int32_t *src,
float mul, int len);
void ff_int32_to_float_fmul_array8_vfp(FmtConvertContext *c, float *dst,
const int32_t *src, const float *mul,
int len);
av_cold void ff_fmt_convert_init_arm(FmtConvertContext *c)
{
@@ -42,11 +36,9 @@ av_cold void ff_fmt_convert_init_arm(FmtConvertContext *c)
if (have_vfp_vm(cpu_flags)) {
c->int32_to_float_fmul_scalar = ff_int32_to_float_fmul_scalar_vfp;
c->int32_to_float_fmul_array8 = ff_int32_to_float_fmul_array8_vfp;
}
if (have_neon(cpu_flags)) {
c->int32_to_float_fmul_array8 = ff_int32_to_float_fmul_array8_neon;
c->int32_to_float_fmul_scalar = ff_int32_to_float_fmul_scalar_neon;
}
}

36
libavcodec/arm/fmtconvert_neon.S
View File

@@ -50,39 +50,3 @@ NOVFP len .req r3
bx lr
.unreq len
endfunc
function ff_int32_to_float_fmul_array8_neon, export=1
ldr r0, [sp]
lsr r0, r0, #3
subs r0, r0, #1
beq 1f
2:
vld1.32 {q0-q1}, [r2,:128]!
vld1.32 {q2-q3}, [r2,:128]!
vld1.32 {d20}, [r3]!
subs r0, r0, #2
vcvt.f32.s32 q0, q0
vcvt.f32.s32 q1, q1
vdup.32 q8, d20[0]
vcvt.f32.s32 q2, q2
vcvt.f32.s32 q3, q3
vmul.f32 q0, q0, q8
vdup.32 q9, d20[1]
vmul.f32 q1, q1, q8
vmul.f32 q2, q2, q9
vmul.f32 q3, q3, q9
vst1.32 {q0-q1}, [r1,:128]!
vst1.32 {q2-q3}, [r1,:128]!
bgt 2b
it lt
bxlt lr
1:
vld1.32 {q0-q1}, [r2,:128]
vld1.32 {d16[],d17[]}, [r3]
vcvt.f32.s32 q0, q0
vcvt.f32.s32 q1, q1
vmul.f32 q0, q0, q8
vmul.f32 q1, q1, q8
vst1.32 {q0-q1}, [r1,:128]
bx lr
endfunc

161
libavcodec/arm/fmtconvert_vfp.S
View File

@@ -21,167 +21,6 @@
#include "config.h"
#include "libavutil/arm/asm.S"
/**
* ARM VFP optimised int32 to float conversion.
* Assume len is a multiple of 8, destination buffer is at least 4 bytes aligned
* (16 bytes alignment is best for BCM2835), little-endian.
*/
@ void ff_int32_to_float_fmul_array8_vfp(FmtConvertContext *c, float *dst, const int32_t *src, const float *mul, int len)
function ff_int32_to_float_fmul_array8_vfp, export=1
push {lr}
ldr a1, [sp, #4]
subs lr, a1, #3*8
bcc 50f @ too short to pipeline
@ Now need to find (len / 8) % 3. The approximation
@ x / 24 = (x * 0xAB) >> 12
@ is good for x < 4096, which is true for both AC3 and DCA.
mov a1, #0xAB
ldr ip, =0x03070000 @ RunFast mode, short vectors of length 8, stride 1
mul a1, lr, a1
vpush {s16-s31}
mov a1, a1, lsr #12
add a1, a1, a1, lsl #1
rsb a1, a1, lr, lsr #3
cmp a1, #1
fmrx a1, FPSCR
fmxr FPSCR, ip
beq 11f
blo 10f
@ Array is (2 + multiple of 3) x 8 floats long
@ drop through...
vldmia a3!, {s16-s23}
vldmia a4!, {s2,s3}
vldmia a3!, {s24-s31}
vcvt.f32.s32 s16, s16
vcvt.f32.s32 s17, s17
vcvt.f32.s32 s18, s18
vcvt.f32.s32 s19, s19
vcvt.f32.s32 s20, s20
vcvt.f32.s32 s21, s21
vcvt.f32.s32 s22, s22
vcvt.f32.s32 s23, s23
vmul.f32 s16, s16, s2
@ drop through...
3:
vldmia a3!, {s8-s15}
vldmia a4!, {s1}
vcvt.f32.s32 s24, s24
vcvt.f32.s32 s25, s25
vcvt.f32.s32 s26, s26
vcvt.f32.s32 s27, s27
vcvt.f32.s32 s28, s28
vcvt.f32.s32 s29, s29
vcvt.f32.s32 s30, s30
vcvt.f32.s32 s31, s31
vmul.f32 s24, s24, s3
vstmia a2!, {s16-s19}
vstmia a2!, {s20-s23}
2:
vldmia a3!, {s16-s23}
vldmia a4!, {s2}
vcvt.f32.s32 s8, s8
vcvt.f32.s32 s9, s9
vcvt.f32.s32 s10, s10
vcvt.f32.s32 s11, s11
vcvt.f32.s32 s12, s12
vcvt.f32.s32 s13, s13
vcvt.f32.s32 s14, s14
vcvt.f32.s32 s15, s15
vmul.f32 s8, s8, s1
vstmia a2!, {s24-s27}
vstmia a2!, {s28-s31}
1:
vldmia a3!, {s24-s31}
vldmia a4!, {s3}
vcvt.f32.s32 s16, s16
vcvt.f32.s32 s17, s17
vcvt.f32.s32 s18, s18
vcvt.f32.s32 s19, s19
vcvt.f32.s32 s20, s20
vcvt.f32.s32 s21, s21
vcvt.f32.s32 s22, s22
vcvt.f32.s32 s23, s23
vmul.f32 s16, s16, s2
vstmia a2!, {s8-s11}
vstmia a2!, {s12-s15}
subs lr, lr, #8*3
bpl 3b
vcvt.f32.s32 s24, s24
vcvt.f32.s32 s25, s25
vcvt.f32.s32 s26, s26
vcvt.f32.s32 s27, s27
vcvt.f32.s32 s28, s28
vcvt.f32.s32 s29, s29
vcvt.f32.s32 s30, s30
vcvt.f32.s32 s31, s31
vmul.f32 s24, s24, s3
vstmia a2!, {s16-s19}
vstmia a2!, {s20-s23}
vstmia a2!, {s24-s27}
vstmia a2!, {s28-s31}
fmxr FPSCR, a1
vpop {s16-s31}
pop {pc}
10: @ Array is (multiple of 3) x 8 floats long
vldmia a3!, {s8-s15}
vldmia a4!, {s1,s2}
vldmia a3!, {s16-s23}
vcvt.f32.s32 s8, s8
vcvt.f32.s32 s9, s9
vcvt.f32.s32 s10, s10
vcvt.f32.s32 s11, s11
vcvt.f32.s32 s12, s12
vcvt.f32.s32 s13, s13
vcvt.f32.s32 s14, s14
vcvt.f32.s32 s15, s15
vmul.f32 s8, s8, s1
b 1b
11: @ Array is (1 + multiple of 3) x 8 floats long
vldmia a3!, {s24-s31}
vldmia a4!, {s3}
vldmia a3!, {s8-s15}
vldmia a4!, {s1}
vcvt.f32.s32 s24, s24
vcvt.f32.s32 s25, s25
vcvt.f32.s32 s26, s26
vcvt.f32.s32 s27, s27
vcvt.f32.s32 s28, s28
vcvt.f32.s32 s29, s29
vcvt.f32.s32 s30, s30
vcvt.f32.s32 s31, s31
vmul.f32 s24, s24, s3
b 2b
50:
ldr lr, =0x03070000 @ RunFast mode, short vectors of length 8, stride 1
fmrx ip, FPSCR
fmxr FPSCR, lr
51:
vldmia a3!, {s8-s15}
vldmia a4!, {s0}
vcvt.f32.s32 s8, s8
vcvt.f32.s32 s9, s9
vcvt.f32.s32 s10, s10
vcvt.f32.s32 s11, s11
vcvt.f32.s32 s12, s12
vcvt.f32.s32 s13, s13
vcvt.f32.s32 s14, s14
vcvt.f32.s32 s15, s15
vmul.f32 s8, s8, s0
subs a1, a1, #8
vstmia a2!, {s8-s11}
vstmia a2!, {s12-s15}
bne 51b
fmxr FPSCR, ip
pop {pc}
endfunc
/**
* ARM VFP optimised int32 to float conversion.
* Assume len is a multiple of 8, destination buffer is at least 4 bytes aligned

10
libavcodec/fmtconvert.c
View File

@@ -32,19 +32,9 @@ static void int32_to_float_fmul_scalar_c(float *dst, const int32_t *src,
dst[i] = src[i] * mul;
}
static void int32_to_float_fmul_array8_c(FmtConvertContext *c, float *dst,
const int32_t *src, const float *mul,
int len)
{
int i;
for (i = 0; i < len; i += 8)
c->int32_to_float_fmul_scalar(&dst[i], &src[i], *mul++, 8);
}
av_cold void ff_fmt_convert_init(FmtConvertContext *c)
{
c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_c;
c->int32_to_float_fmul_array8 = int32_to_float_fmul_array8_c;
#if ARCH_AARCH64
ff_fmt_convert_init_aarch64(c);

17
libavcodec/fmtconvert.h
View File

@@ -37,23 +37,6 @@ typedef struct FmtConvertContext {
*/
void (*int32_to_float_fmul_scalar)(float *dst, const int32_t *src,
float mul, int len);
/**
* Convert an array of int32_t to float and multiply by a float value from another array,
* stepping along the float array once for each 8 integers.
* @param c pointer to FmtConvertContext.
* @param dst destination array of float.
* constraints: 16-byte aligned
* @param src source array of int32_t.
* constraints: 16-byte aligned
* @param mul source array of float multipliers.
* @param len number of elements to convert.
* constraints: multiple of 8
*/
void (*int32_to_float_fmul_array8)(struct FmtConvertContext *c,
float *dst, const int32_t *src,
const float *mul, int len);
} FmtConvertContext;
void ff_fmt_convert_init(FmtConvertContext *c);

4
libavcodec/riscv/fmtconvert_init.c
View File

@@ -27,9 +27,6 @@
void ff_int32_to_float_fmul_scalar_rvv(float *dst, const int32_t *src,
float mul, int len);
void ff_int32_to_float_fmul_array8_rvv(FmtConvertContext *c, float *dst,
const int32_t *src, const float *mul,
int len);
av_cold void ff_fmt_convert_init_riscv(FmtConvertContext *c)
{
@@ -38,7 +35,6 @@ av_cold void ff_fmt_convert_init_riscv(FmtConvertContext *c)
if ((flags & AV_CPU_FLAG_RVV_F32) && (flags & AV_CPU_FLAG_RVB)) {
c->int32_to_float_fmul_scalar = ff_int32_to_float_fmul_scalar_rvv;
c->int32_to_float_fmul_array8 = ff_int32_to_float_fmul_array8_rvv;
}
#endif
}

29
libavcodec/riscv/fmtconvert_rvv.S
View File

@@ -37,32 +37,3 @@ NOHWF mv a2, a3
ret
endfunc
func ff_int32_to_float_fmul_array8_rvv, zve32f, zba
lpad 0
srai a4, a4, 3
1: vsetvli t0, a4, e32, m1, ta, ma
vle32.v v24, (a3)
slli t2, t0, 2 + 3
vlseg8e32.v v16, (a2)
vsetvli t3, zero, e32, m8, ta, ma
vfcvt.f.x.v v16, v16
vsetvli zero, a4, e32, m1, ta, ma
vfmul.vv v16, v16, v24
sub a4, a4, t0
vfmul.vv v17, v17, v24
sh2add a3, t0, a3
vfmul.vv v18, v18, v24
add a2, a2, t2
vfmul.vv v19, v19, v24
vfmul.vv v20, v20, v24
vfmul.vv v21, v21, v24
vfmul.vv v22, v22, v24
vfmul.vv v23, v23, v24
vsseg8e32.v v16, (a1)
add a1, a1, t2
bnez a4, 1b
ret
endfunc

28
libavcodec/x86/fmtconvert.asm
View File

@@ -57,31 +57,3 @@ cglobal int32_to_float_fmul_scalar, 4, 4, %1, dst, src, mul, len
INIT_XMM sse2
INT32_TO_FLOAT_FMUL_SCALAR 3
;------------------------------------------------------------------------------
; void ff_int32_to_float_fmul_array8(FmtConvertContext *c, float *dst, const int32_t *src,
; const float *mul, int len);
;------------------------------------------------------------------------------
%macro INT32_TO_FLOAT_FMUL_ARRAY8 0
cglobal int32_to_float_fmul_array8, 5, 5, 5, c, dst, src, mul, len
shl lend, 2
add srcq, lenq
add dstq, lenq
neg lenq
.loop:
movss m0, [mulq]
SPLATD m0
cvtdq2ps m1, [srcq+lenq ]
cvtdq2ps m2, [srcq+lenq+16]
mulps m1, m0
mulps m2, m0
mova [dstq+lenq ], m1
mova [dstq+lenq+16], m2
add mulq, 4
add lenq, 32
jl .loop
RET
%endmacro
INIT_XMM sse2
INT32_TO_FLOAT_FMUL_ARRAY8

3
libavcodec/x86/fmtconvert_init.c
View File

@@ -28,8 +28,6 @@
#include "libavcodec/fmtconvert.h"
void ff_int32_to_float_fmul_scalar_sse2(float *dst, const int32_t *src, float mul, int len);
void ff_int32_to_float_fmul_array8_sse2(FmtConvertContext *c, float *dst, const int32_t *src,
const float *mul, int len);
av_cold void ff_fmt_convert_init_x86(FmtConvertContext *c)
{
@@ -37,6 +35,5 @@ av_cold void ff_fmt_convert_init_x86(FmtConvertContext *c)
if (EXTERNAL_SSE2(cpu_flags)) {
c->int32_to_float_fmul_scalar = ff_int32_to_float_fmul_scalar_sse2;
c->int32_to_float_fmul_array8 = ff_int32_to_float_fmul_array8_sse2;
}
}

22
tests/checkasm/fmtconvert.c
View File

@@ -81,27 +81,5 @@ void checkasm_check_fmtconvert(void)
}
}
}
if (check_func(c.int32_to_float_fmul_array8, "int32_to_float_fmul_array8")) {
declare_func(void, FmtConvertContext *, float *, const int32_t *,
const float *, int);
for (i = 0; i < 4; i++) {
for (j = 0; j < FF_ARRAY_ELEMS(length); j++) {
randomize_input(length[j]);
call_ref(&c, dst0, in, scale_arr, length[j]);
call_new(&c, dst1, in, scale_arr, length[j]);
if (!float_near_ulp_array(dst0, dst1, 3, length[j])) {
fail();
fprintf(stderr, "int32_to_float_fmul_array8: len: %d\n", length[j]);
break;
}
bench_new(&c, dst1, in, scale_arr, length[j]);
}
}
}
report("fmtconvert");
}