From dd2940113739833eeb75be1931a2e3f74ae4743d Mon Sep 17 00:00:00 2001 From: Nigel Tao Date: Wed, 28 Jun 2023 13:49:55 +1000 Subject: [PATCH] std/jpeg: re-format generated IDCT code --- script/print-jpeg-idct-code.go | 107 +++-- std/jpeg/decode_idct_default.wuffs | 672 ++++++++++++----------------- 2 files changed, 329 insertions(+), 450 deletions(-) diff --git a/script/print-jpeg-idct-code.go b/script/print-jpeg-idct-code.go index 8a2c6201..22bed641 100644 --- a/script/print-jpeg-idct-code.go +++ b/script/print-jpeg-idct-code.go @@ -38,31 +38,31 @@ func main1() error { fmt.Printf("// -------- BEGIN generated by script/print-jpeg-idct-code.go\n") fmt.Println() - fmt.Printf("// p0_298631336 = 0x%04X = %5d\n", round(p0_298631336), round(p0_298631336)) - fmt.Printf("// p0_390180644 = 0x%04X = %5d\n", round(p0_390180644), round(p0_390180644)) - fmt.Printf("// p0_509795579 = 0x%04X = %5d\n", round(p0_509795579), round(p0_509795579)) - fmt.Printf("// p0_541196100 = 0x%04X = %5d\n", round(p0_541196100), round(p0_541196100)) - fmt.Printf("// p0_601344887 = 0x%04X = %5d\n", round(p0_601344887), round(p0_601344887)) - fmt.Printf("// p0_765366865 = 0x%04X = %5d\n", round(p0_765366865), round(p0_765366865)) - fmt.Printf("// p0_785694958 = 0x%04X = %5d\n", round(p0_785694958), round(p0_785694958)) - fmt.Printf("// p0_899976223 = 0x%04X = %5d\n", round(p0_899976223), round(p0_899976223)) - fmt.Printf("// p1_175875602 = 0x%04X = %5d\n", round(p1_175875602), round(p1_175875602)) - fmt.Printf("// p1_306562965 = 0x%04X = %5d\n", round(p1_306562965), round(p1_306562965)) - fmt.Printf("// p1_501321110 = 0x%04X = %5d\n", round(p1_501321110), round(p1_501321110)) - fmt.Printf("// p1_847759065 = 0x%04X = %5d\n", round(p1_847759065), round(p1_847759065)) - fmt.Printf("// p1_961570560 = 0x%04X = %5d\n", round(p1_961570560), round(p1_961570560)) - fmt.Printf("// p2_053119869 = 0x%04X = %5d\n", round(p2_053119869), round(p2_053119869)) - fmt.Printf("// p2_562915447 = 0x%04X = %5d\n", round(p2_562915447), round(p2_562915447)) - fmt.Printf("// p3_072711026 = 0x%04X = %5d\n", round(p3_072711026), round(p3_072711026)) + fmt.Printf("// p0_298631336 = %s = %10d\n", format8X(p0_298631336), round(p0_298631336)) + fmt.Printf("// p0_390180644 = %s = %10d\n", format8X(p0_390180644), round(p0_390180644)) + fmt.Printf("// p0_509795579 = %s = %10d\n", format8X(p0_509795579), round(p0_509795579)) + fmt.Printf("// p0_541196100 = %s = %10d\n", format8X(p0_541196100), round(p0_541196100)) + fmt.Printf("// p0_601344887 = %s = %10d\n", format8X(p0_601344887), round(p0_601344887)) + fmt.Printf("// p0_765366865 = %s = %10d\n", format8X(p0_765366865), round(p0_765366865)) + fmt.Printf("// p0_785694958 = %s = %10d\n", format8X(p0_785694958), round(p0_785694958)) + fmt.Printf("// p0_899976223 = %s = %10d\n", format8X(p0_899976223), round(p0_899976223)) + fmt.Printf("// p1_175875602 = %s = %10d\n", format8X(p1_175875602), round(p1_175875602)) + fmt.Printf("// p1_306562965 = %s = %10d\n", format8X(p1_306562965), round(p1_306562965)) + fmt.Printf("// p1_501321110 = %s = %10d\n", format8X(p1_501321110), round(p1_501321110)) + fmt.Printf("// p1_847759065 = %s = %10d\n", format8X(p1_847759065), round(p1_847759065)) + fmt.Printf("// p1_961570560 = %s = %10d\n", format8X(p1_961570560), round(p1_961570560)) + fmt.Printf("// p2_053119869 = %s = %10d\n", format8X(p2_053119869), round(p2_053119869)) + fmt.Printf("// p2_562915447 = %s = %10d\n", format8X(p2_562915447), round(p2_562915447)) + fmt.Printf("// p3_072711026 = %s = %10d\n", format8X(p3_072711026), round(p3_072711026)) fmt.Printf("//\n") - fmt.Printf("// m0_390180644 = 0x%08X = %10d\n", round(m0_390180644), round(m0_390180644)) - fmt.Printf("// m0_509795579 = 0x%08X = %10d\n", round(m0_509795579), round(m0_509795579)) - fmt.Printf("// m0_601344887 = 0x%08X = %10d\n", round(m0_601344887), round(m0_601344887)) - fmt.Printf("// m0_785694958 = 0x%08X = %10d\n", round(m0_785694958), round(m0_785694958)) - fmt.Printf("// m0_899976223 = 0x%08X = %10d\n", round(m0_899976223), round(m0_899976223)) - fmt.Printf("// m1_306562965 = 0x%08X = %10d\n", round(m1_306562965), round(m1_306562965)) - fmt.Printf("// m1_961570560 = 0x%08X = %10d\n", round(m1_961570560), round(m1_961570560)) - fmt.Printf("// m2_562915447 = 0x%08X = %10d\n", round(m2_562915447), round(m2_562915447)) + fmt.Printf("// m0_390180644 = %s = %10d\n", format8X(m0_390180644), round(m0_390180644)) + fmt.Printf("// m0_509795579 = %s = %10d\n", format8X(m0_509795579), round(m0_509795579)) + fmt.Printf("// m0_601344887 = %s = %10d\n", format8X(m0_601344887), round(m0_601344887)) + fmt.Printf("// m0_785694958 = %s = %10d\n", format8X(m0_785694958), round(m0_785694958)) + fmt.Printf("// m0_899976223 = %s = %10d\n", format8X(m0_899976223), round(m0_899976223)) + fmt.Printf("// m1_306562965 = %s = %10d\n", format8X(m1_306562965), round(m1_306562965)) + fmt.Printf("// m1_961570560 = %s = %10d\n", format8X(m1_961570560), round(m1_961570560)) + fmt.Printf("// m2_562915447 = %s = %10d\n", format8X(m2_562915447), round(m2_562915447)) fmt.Println() for x := 0; x < 8; x++ { @@ -109,31 +109,36 @@ func replace(s string, m map[string]string) string { } func addConsts(m map[string]string) map[string]string { - m["$p0_298631336$"] = fmt.Sprintf("0x%04X", round(p0_298631336)) - m["$p0_509795579$"] = fmt.Sprintf("0x%04X", round(p0_509795579)) - m["$p0_541196100$"] = fmt.Sprintf("0x%04X", round(p0_541196100)) - m["$p0_601344887$"] = fmt.Sprintf("0x%04X", round(p0_601344887)) - m["$p0_765366865$"] = fmt.Sprintf("0x%04X", round(p0_765366865)) - m["$p0_785694958$"] = fmt.Sprintf("0x%04X", round(p0_785694958)) - m["$p1_175875602$"] = fmt.Sprintf("0x%04X", round(p1_175875602)) - m["$p1_306562965$"] = fmt.Sprintf("0x%04X", round(p1_306562965)) - m["$p1_501321110$"] = fmt.Sprintf("0x%04X", round(p1_501321110)) - m["$p1_847759065$"] = fmt.Sprintf("0x%04X", round(p1_847759065)) - m["$p2_053119869$"] = fmt.Sprintf("0x%04X", round(p2_053119869)) - m["$p3_072711026$"] = fmt.Sprintf("0x%04X", round(p3_072711026)) + m["$p0_298631336$"] = format8X(p0_298631336) + m["$p0_509795579$"] = format8X(p0_509795579) + m["$p0_541196100$"] = format8X(p0_541196100) + m["$p0_601344887$"] = format8X(p0_601344887) + m["$p0_765366865$"] = format8X(p0_765366865) + m["$p0_785694958$"] = format8X(p0_785694958) + m["$p1_175875602$"] = format8X(p1_175875602) + m["$p1_306562965$"] = format8X(p1_306562965) + m["$p1_501321110$"] = format8X(p1_501321110) + m["$p1_847759065$"] = format8X(p1_847759065) + m["$p2_053119869$"] = format8X(p2_053119869) + m["$p3_072711026$"] = format8X(p3_072711026) - m["$m0_390180644$"] = fmt.Sprintf("0x%08X", round(m0_390180644)) - m["$m0_509795579$"] = fmt.Sprintf("0x%08X", round(m0_509795579)) - m["$m0_601344887$"] = fmt.Sprintf("0x%08X", round(m0_601344887)) - m["$m0_785694958$"] = fmt.Sprintf("0x%08X", round(m0_785694958)) - m["$m0_899976223$"] = fmt.Sprintf("0x%08X", round(m0_899976223)) - m["$m1_306562965$"] = fmt.Sprintf("0x%08X", round(m1_306562965)) - m["$m1_961570560$"] = fmt.Sprintf("0x%08X", round(m1_961570560)) - m["$m2_562915447$"] = fmt.Sprintf("0x%08X", round(m2_562915447)) + m["$m0_390180644$"] = format8X(m0_390180644) + m["$m0_509795579$"] = format8X(m0_509795579) + m["$m0_601344887$"] = format8X(m0_601344887) + m["$m0_785694958$"] = format8X(m0_785694958) + m["$m0_899976223$"] = format8X(m0_899976223) + m["$m1_306562965$"] = format8X(m1_306562965) + m["$m1_961570560$"] = format8X(m1_961570560) + m["$m2_562915447$"] = format8X(m2_562915447) return m } +func format8X(x float64) string { + s := fmt.Sprintf("0x%08X", round(x)) + return s[:6] + "_" + s[6:] +} + func round(x float64) uint32 { // Hard-code some rounding errors to match libjpeg-turbo exactly. // @@ -257,10 +262,8 @@ bq6 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][$row6colX$]) ~ // This code... ca = (bq2 ~mod+ bq6) ~mod* $p0_541196100$ - cb2 = ca ~mod+ (bq2 ~mod* $p0_765366865$) cb6 = ca ~mod- (bq6 ~mod* $p1_847759065$) - // ...is equivalent to this more-SIMD-like code. // // cb2 = (bq2 ~mod* $p1_306562965$) ~mod+ (bq6 ~mod* $p0_541196100$) @@ -291,25 +294,20 @@ ci73 = bq7 ~mod+ bq3 // This code... cj = (ci73 ~mod+ ci51) ~mod* $p1_175875602$ - ck1 = bq1 ~mod* $p1_501321110$ ck3 = bq3 ~mod* $p3_072711026$ ck5 = bq5 ~mod* $p2_053119869$ ck7 = bq7 ~mod* $p0_298631336$ - ci51 ~mod*= $m0_390180644$ ci53 ~mod*= $m2_562915447$ ci71 ~mod*= $m0_899976223$ ci73 ~mod*= $m1_961570560$ - cl51 = ci51 ~mod+ cj cl73 = ci73 ~mod+ cj - ck1 ~mod+= ci71 ~mod+ cl51 ck3 ~mod+= ci53 ~mod+ cl73 ck5 ~mod+= ci53 ~mod+ cl51 ck7 ~mod+= ci71 ~mod+ cl73 - // ...is equivalent to this more-SIMD-like code. // // cl73 = (ci73 ~mod* $m0_785694958$) ~mod+ (ci51 ~mod* $p1_175875602$) @@ -365,10 +363,8 @@ in6 = intermediate[$rowYcol6$] // This code... ra = (in2 ~mod+ in6) ~mod* $p0_541196100$ - rb2 = ra ~mod+ (in2 ~mod* $p0_765366865$) rb6 = ra ~mod- (in6 ~mod* $p1_847759065$) - // ...is equivalent to this more-SIMD-like code. // // rb2 = (in2 ~mod* $p1_306562965$) ~mod+ (in6 ~mod* $p0_541196100$) @@ -399,25 +395,20 @@ ri73 = in7 ~mod+ in3 // This code... rj = (ri73 ~mod+ ri51) ~mod* $p1_175875602$ - rk1 = in1 ~mod* $p1_501321110$ rk3 = in3 ~mod* $p3_072711026$ rk5 = in5 ~mod* $p2_053119869$ rk7 = in7 ~mod* $p0_298631336$ - ri51 ~mod*= $m0_390180644$ ri53 ~mod*= $m2_562915447$ ri71 ~mod*= $m0_899976223$ ri73 ~mod*= $m1_961570560$ - rl51 = ri51 ~mod+ rj rl73 = ri73 ~mod+ rj - rk1 ~mod+= ri71 ~mod+ rl51 rk3 ~mod+= ri53 ~mod+ rl73 rk5 ~mod+= ri53 ~mod+ rl51 rk7 ~mod+= ri71 ~mod+ rl73 - // ...is equivalent to this more-SIMD-like code. // // rl73 = (ri73 ~mod* $m0_785694958$) ~mod+ (ri51 ~mod* $p1_175875602$) diff --git a/std/jpeg/decode_idct_default.wuffs b/std/jpeg/decode_idct_default.wuffs index a2a20df3..38b81fa1 100644 --- a/std/jpeg/decode_idct_default.wuffs +++ b/std/jpeg/decode_idct_default.wuffs @@ -99,31 +99,31 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q // -------- BEGIN generated by script/print-jpeg-idct-code.go - // p0_298631336 = 0x098E = 2446 - // p0_390180644 = 0x0C7C = 3196 - // p0_509795579 = 0x1051 = 4177 - // p0_541196100 = 0x1151 = 4433 - // p0_601344887 = 0x133E = 4926 - // p0_765366865 = 0x187E = 6270 - // p0_785694958 = 0x1925 = 6437 - // p0_899976223 = 0x1CCD = 7373 - // p1_175875602 = 0x25A1 = 9633 - // p1_306562965 = 0x29CF = 10703 - // p1_501321110 = 0x300B = 12299 - // p1_847759065 = 0x3B21 = 15137 - // p1_961570560 = 0x3EC5 = 16069 - // p2_053119869 = 0x41B3 = 16819 - // p2_562915447 = 0x5203 = 20995 - // p3_072711026 = 0x6254 = 25172 + // p0_298631336 = 0x0000_098E = 2446 + // p0_390180644 = 0x0000_0C7C = 3196 + // p0_509795579 = 0x0000_1051 = 4177 + // p0_541196100 = 0x0000_1151 = 4433 + // p0_601344887 = 0x0000_133E = 4926 + // p0_765366865 = 0x0000_187E = 6270 + // p0_785694958 = 0x0000_1925 = 6437 + // p0_899976223 = 0x0000_1CCD = 7373 + // p1_175875602 = 0x0000_25A1 = 9633 + // p1_306562965 = 0x0000_29CF = 10703 + // p1_501321110 = 0x0000_300B = 12299 + // p1_847759065 = 0x0000_3B21 = 15137 + // p1_961570560 = 0x0000_3EC5 = 16069 + // p2_053119869 = 0x0000_41B3 = 16819 + // p2_562915447 = 0x0000_5203 = 20995 + // p3_072711026 = 0x0000_6254 = 25172 // - // m0_390180644 = 0xFFFFF384 = 4294964100 - // m0_509795579 = 0xFFFFEFB0 = 4294963120 - // m0_601344887 = 0xFFFFECC1 = 4294962369 - // m0_785694958 = 0xFFFFE6DC = 4294960860 - // m0_899976223 = 0xFFFFE333 = 4294959923 - // m1_306562965 = 0xFFFFD630 = 4294956592 - // m1_961570560 = 0xFFFFC13B = 4294951227 - // m2_562915447 = 0xFFFFADFD = 4294946301 + // m0_390180644 = 0xFFFF_F384 = 4294964100 + // m0_509795579 = 0xFFFF_EFB0 = 4294963120 + // m0_601344887 = 0xFFFF_ECC1 = 4294962369 + // m0_785694958 = 0xFFFF_E6DC = 4294960860 + // m0_899976223 = 0xFFFF_E333 = 4294959923 + // m1_306562965 = 0xFFFF_D630 = 4294956592 + // m1_961570560 = 0xFFFF_C13B = 4294951227 + // m2_562915447 = 0xFFFF_ADFD = 4294946301 // ==== First pass, column 0. @@ -155,15 +155,13 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q bq6 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x30]) ~mod* (this.quant_tables[args.q][0x30] as base.u32) // This code... - ca = (bq2 ~mod+ bq6) ~mod* 0x1151 - - cb2 = ca ~mod+ (bq2 ~mod* 0x187E) - cb6 = ca ~mod- (bq6 ~mod* 0x3B21) - + ca = (bq2 ~mod+ bq6) ~mod* 0x0000_1151 + cb2 = ca ~mod+ (bq2 ~mod* 0x0000_187E) + cb6 = ca ~mod- (bq6 ~mod* 0x0000_3B21) // ...is equivalent to this more-SIMD-like code. // - // cb2 = (bq2 ~mod* 0x29CF) ~mod+ (bq6 ~mod* 0x1151) - // cb6 = (bq2 ~mod* 0x1151) ~mod+ (bq6 ~mod* 0xFFFFD630) + // cb2 = (bq2 ~mod* 0x0000_29CF) ~mod+ (bq6 ~mod* 0x0000_1151) + // cb6 = (bq2 ~mod* 0x0000_1151) ~mod+ (bq6 ~mod* 0xFFFF_D630) bq0 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x00]) ~mod* (this.quant_tables[args.q][0x00] as base.u32) bq4 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x20]) ~mod* (this.quant_tables[args.q][0x20] as base.u32) @@ -189,34 +187,29 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q ci73 = bq7 ~mod+ bq3 // This code... - cj = (ci73 ~mod+ ci51) ~mod* 0x25A1 - - ck1 = bq1 ~mod* 0x300B - ck3 = bq3 ~mod* 0x6254 - ck5 = bq5 ~mod* 0x41B3 - ck7 = bq7 ~mod* 0x098E - + cj = (ci73 ~mod+ ci51) ~mod* 0x0000_25A1 + ck1 = bq1 ~mod* 0x0000_300B + ck3 = bq3 ~mod* 0x0000_6254 + ck5 = bq5 ~mod* 0x0000_41B3 + ck7 = bq7 ~mod* 0x0000_098E ci51 ~mod*= 0xFFFF_F384 ci53 ~mod*= 0xFFFF_ADFD ci71 ~mod*= 0xFFFF_E333 ci73 ~mod*= 0xFFFF_C13B - cl51 = ci51 ~mod+ cj cl73 = ci73 ~mod+ cj - ck1 ~mod+= ci71 ~mod+ cl51 ck3 ~mod+= ci53 ~mod+ cl73 ck5 ~mod+= ci53 ~mod+ cl51 ck7 ~mod+= ci71 ~mod+ cl73 - // ...is equivalent to this more-SIMD-like code. // - // cl73 = (ci73 ~mod* 0xFFFFE6DC) ~mod+ (ci51 ~mod* 0x25A1) - // cl51 = (ci73 ~mod* 0x25A1) ~mod+ (ci51 ~mod* 0x1925) - // ck1 = cl51 ~mod+ ((bq1 ~mod* 0x133E) ~mod+ (bq7 ~mod* 0xFFFFE333)) - // ck3 = cl73 ~mod+ ((bq3 ~mod* 0x1051) ~mod+ (bq5 ~mod* 0xFFFFADFD)) - // ck5 = cl51 ~mod+ ((bq3 ~mod* 0xFFFFADFD) ~mod+ (bq5 ~mod* 0xFFFFEFB0)) - // ck7 = cl73 ~mod+ ((bq1 ~mod* 0xFFFFE333) ~mod+ (bq7 ~mod* 0xFFFFECC1)) + // cl73 = (ci73 ~mod* 0xFFFF_E6DC) ~mod+ (ci51 ~mod* 0x0000_25A1) + // cl51 = (ci73 ~mod* 0x0000_25A1) ~mod+ (ci51 ~mod* 0x0000_1925) + // ck1 = cl51 ~mod+ ((bq1 ~mod* 0x0000_133E) ~mod+ (bq7 ~mod* 0xFFFF_E333)) + // ck3 = cl73 ~mod+ ((bq3 ~mod* 0x0000_1051) ~mod+ (bq5 ~mod* 0xFFFF_ADFD)) + // ck5 = cl51 ~mod+ ((bq3 ~mod* 0xFFFF_ADFD) ~mod+ (bq5 ~mod* 0xFFFF_EFB0)) + // ck7 = cl73 ~mod+ ((bq1 ~mod* 0xFFFF_E333) ~mod+ (bq7 ~mod* 0xFFFF_ECC1)) // Combine rows. @@ -260,15 +253,13 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q bq6 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x31]) ~mod* (this.quant_tables[args.q][0x31] as base.u32) // This code... - ca = (bq2 ~mod+ bq6) ~mod* 0x1151 - - cb2 = ca ~mod+ (bq2 ~mod* 0x187E) - cb6 = ca ~mod- (bq6 ~mod* 0x3B21) - + ca = (bq2 ~mod+ bq6) ~mod* 0x0000_1151 + cb2 = ca ~mod+ (bq2 ~mod* 0x0000_187E) + cb6 = ca ~mod- (bq6 ~mod* 0x0000_3B21) // ...is equivalent to this more-SIMD-like code. // - // cb2 = (bq2 ~mod* 0x29CF) ~mod+ (bq6 ~mod* 0x1151) - // cb6 = (bq2 ~mod* 0x1151) ~mod+ (bq6 ~mod* 0xFFFFD630) + // cb2 = (bq2 ~mod* 0x0000_29CF) ~mod+ (bq6 ~mod* 0x0000_1151) + // cb6 = (bq2 ~mod* 0x0000_1151) ~mod+ (bq6 ~mod* 0xFFFF_D630) bq0 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x01]) ~mod* (this.quant_tables[args.q][0x01] as base.u32) bq4 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x21]) ~mod* (this.quant_tables[args.q][0x21] as base.u32) @@ -294,34 +285,29 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q ci73 = bq7 ~mod+ bq3 // This code... - cj = (ci73 ~mod+ ci51) ~mod* 0x25A1 - - ck1 = bq1 ~mod* 0x300B - ck3 = bq3 ~mod* 0x6254 - ck5 = bq5 ~mod* 0x41B3 - ck7 = bq7 ~mod* 0x098E - + cj = (ci73 ~mod+ ci51) ~mod* 0x0000_25A1 + ck1 = bq1 ~mod* 0x0000_300B + ck3 = bq3 ~mod* 0x0000_6254 + ck5 = bq5 ~mod* 0x0000_41B3 + ck7 = bq7 ~mod* 0x0000_098E ci51 ~mod*= 0xFFFF_F384 ci53 ~mod*= 0xFFFF_ADFD ci71 ~mod*= 0xFFFF_E333 ci73 ~mod*= 0xFFFF_C13B - cl51 = ci51 ~mod+ cj cl73 = ci73 ~mod+ cj - ck1 ~mod+= ci71 ~mod+ cl51 ck3 ~mod+= ci53 ~mod+ cl73 ck5 ~mod+= ci53 ~mod+ cl51 ck7 ~mod+= ci71 ~mod+ cl73 - // ...is equivalent to this more-SIMD-like code. // - // cl73 = (ci73 ~mod* 0xFFFFE6DC) ~mod+ (ci51 ~mod* 0x25A1) - // cl51 = (ci73 ~mod* 0x25A1) ~mod+ (ci51 ~mod* 0x1925) - // ck1 = cl51 ~mod+ ((bq1 ~mod* 0x133E) ~mod+ (bq7 ~mod* 0xFFFFE333)) - // ck3 = cl73 ~mod+ ((bq3 ~mod* 0x1051) ~mod+ (bq5 ~mod* 0xFFFFADFD)) - // ck5 = cl51 ~mod+ ((bq3 ~mod* 0xFFFFADFD) ~mod+ (bq5 ~mod* 0xFFFFEFB0)) - // ck7 = cl73 ~mod+ ((bq1 ~mod* 0xFFFFE333) ~mod+ (bq7 ~mod* 0xFFFFECC1)) + // cl73 = (ci73 ~mod* 0xFFFF_E6DC) ~mod+ (ci51 ~mod* 0x0000_25A1) + // cl51 = (ci73 ~mod* 0x0000_25A1) ~mod+ (ci51 ~mod* 0x0000_1925) + // ck1 = cl51 ~mod+ ((bq1 ~mod* 0x0000_133E) ~mod+ (bq7 ~mod* 0xFFFF_E333)) + // ck3 = cl73 ~mod+ ((bq3 ~mod* 0x0000_1051) ~mod+ (bq5 ~mod* 0xFFFF_ADFD)) + // ck5 = cl51 ~mod+ ((bq3 ~mod* 0xFFFF_ADFD) ~mod+ (bq5 ~mod* 0xFFFF_EFB0)) + // ck7 = cl73 ~mod+ ((bq1 ~mod* 0xFFFF_E333) ~mod+ (bq7 ~mod* 0xFFFF_ECC1)) // Combine rows. @@ -365,15 +351,13 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q bq6 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x32]) ~mod* (this.quant_tables[args.q][0x32] as base.u32) // This code... - ca = (bq2 ~mod+ bq6) ~mod* 0x1151 - - cb2 = ca ~mod+ (bq2 ~mod* 0x187E) - cb6 = ca ~mod- (bq6 ~mod* 0x3B21) - + ca = (bq2 ~mod+ bq6) ~mod* 0x0000_1151 + cb2 = ca ~mod+ (bq2 ~mod* 0x0000_187E) + cb6 = ca ~mod- (bq6 ~mod* 0x0000_3B21) // ...is equivalent to this more-SIMD-like code. // - // cb2 = (bq2 ~mod* 0x29CF) ~mod+ (bq6 ~mod* 0x1151) - // cb6 = (bq2 ~mod* 0x1151) ~mod+ (bq6 ~mod* 0xFFFFD630) + // cb2 = (bq2 ~mod* 0x0000_29CF) ~mod+ (bq6 ~mod* 0x0000_1151) + // cb6 = (bq2 ~mod* 0x0000_1151) ~mod+ (bq6 ~mod* 0xFFFF_D630) bq0 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x02]) ~mod* (this.quant_tables[args.q][0x02] as base.u32) bq4 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x22]) ~mod* (this.quant_tables[args.q][0x22] as base.u32) @@ -399,34 +383,29 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q ci73 = bq7 ~mod+ bq3 // This code... - cj = (ci73 ~mod+ ci51) ~mod* 0x25A1 - - ck1 = bq1 ~mod* 0x300B - ck3 = bq3 ~mod* 0x6254 - ck5 = bq5 ~mod* 0x41B3 - ck7 = bq7 ~mod* 0x098E - + cj = (ci73 ~mod+ ci51) ~mod* 0x0000_25A1 + ck1 = bq1 ~mod* 0x0000_300B + ck3 = bq3 ~mod* 0x0000_6254 + ck5 = bq5 ~mod* 0x0000_41B3 + ck7 = bq7 ~mod* 0x0000_098E ci51 ~mod*= 0xFFFF_F384 ci53 ~mod*= 0xFFFF_ADFD ci71 ~mod*= 0xFFFF_E333 ci73 ~mod*= 0xFFFF_C13B - cl51 = ci51 ~mod+ cj cl73 = ci73 ~mod+ cj - ck1 ~mod+= ci71 ~mod+ cl51 ck3 ~mod+= ci53 ~mod+ cl73 ck5 ~mod+= ci53 ~mod+ cl51 ck7 ~mod+= ci71 ~mod+ cl73 - // ...is equivalent to this more-SIMD-like code. // - // cl73 = (ci73 ~mod* 0xFFFFE6DC) ~mod+ (ci51 ~mod* 0x25A1) - // cl51 = (ci73 ~mod* 0x25A1) ~mod+ (ci51 ~mod* 0x1925) - // ck1 = cl51 ~mod+ ((bq1 ~mod* 0x133E) ~mod+ (bq7 ~mod* 0xFFFFE333)) - // ck3 = cl73 ~mod+ ((bq3 ~mod* 0x1051) ~mod+ (bq5 ~mod* 0xFFFFADFD)) - // ck5 = cl51 ~mod+ ((bq3 ~mod* 0xFFFFADFD) ~mod+ (bq5 ~mod* 0xFFFFEFB0)) - // ck7 = cl73 ~mod+ ((bq1 ~mod* 0xFFFFE333) ~mod+ (bq7 ~mod* 0xFFFFECC1)) + // cl73 = (ci73 ~mod* 0xFFFF_E6DC) ~mod+ (ci51 ~mod* 0x0000_25A1) + // cl51 = (ci73 ~mod* 0x0000_25A1) ~mod+ (ci51 ~mod* 0x0000_1925) + // ck1 = cl51 ~mod+ ((bq1 ~mod* 0x0000_133E) ~mod+ (bq7 ~mod* 0xFFFF_E333)) + // ck3 = cl73 ~mod+ ((bq3 ~mod* 0x0000_1051) ~mod+ (bq5 ~mod* 0xFFFF_ADFD)) + // ck5 = cl51 ~mod+ ((bq3 ~mod* 0xFFFF_ADFD) ~mod+ (bq5 ~mod* 0xFFFF_EFB0)) + // ck7 = cl73 ~mod+ ((bq1 ~mod* 0xFFFF_E333) ~mod+ (bq7 ~mod* 0xFFFF_ECC1)) // Combine rows. @@ -470,15 +449,13 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q bq6 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x33]) ~mod* (this.quant_tables[args.q][0x33] as base.u32) // This code... - ca = (bq2 ~mod+ bq6) ~mod* 0x1151 - - cb2 = ca ~mod+ (bq2 ~mod* 0x187E) - cb6 = ca ~mod- (bq6 ~mod* 0x3B21) - + ca = (bq2 ~mod+ bq6) ~mod* 0x0000_1151 + cb2 = ca ~mod+ (bq2 ~mod* 0x0000_187E) + cb6 = ca ~mod- (bq6 ~mod* 0x0000_3B21) // ...is equivalent to this more-SIMD-like code. // - // cb2 = (bq2 ~mod* 0x29CF) ~mod+ (bq6 ~mod* 0x1151) - // cb6 = (bq2 ~mod* 0x1151) ~mod+ (bq6 ~mod* 0xFFFFD630) + // cb2 = (bq2 ~mod* 0x0000_29CF) ~mod+ (bq6 ~mod* 0x0000_1151) + // cb6 = (bq2 ~mod* 0x0000_1151) ~mod+ (bq6 ~mod* 0xFFFF_D630) bq0 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x03]) ~mod* (this.quant_tables[args.q][0x03] as base.u32) bq4 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x23]) ~mod* (this.quant_tables[args.q][0x23] as base.u32) @@ -504,34 +481,29 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q ci73 = bq7 ~mod+ bq3 // This code... - cj = (ci73 ~mod+ ci51) ~mod* 0x25A1 - - ck1 = bq1 ~mod* 0x300B - ck3 = bq3 ~mod* 0x6254 - ck5 = bq5 ~mod* 0x41B3 - ck7 = bq7 ~mod* 0x098E - + cj = (ci73 ~mod+ ci51) ~mod* 0x0000_25A1 + ck1 = bq1 ~mod* 0x0000_300B + ck3 = bq3 ~mod* 0x0000_6254 + ck5 = bq5 ~mod* 0x0000_41B3 + ck7 = bq7 ~mod* 0x0000_098E ci51 ~mod*= 0xFFFF_F384 ci53 ~mod*= 0xFFFF_ADFD ci71 ~mod*= 0xFFFF_E333 ci73 ~mod*= 0xFFFF_C13B - cl51 = ci51 ~mod+ cj cl73 = ci73 ~mod+ cj - ck1 ~mod+= ci71 ~mod+ cl51 ck3 ~mod+= ci53 ~mod+ cl73 ck5 ~mod+= ci53 ~mod+ cl51 ck7 ~mod+= ci71 ~mod+ cl73 - // ...is equivalent to this more-SIMD-like code. // - // cl73 = (ci73 ~mod* 0xFFFFE6DC) ~mod+ (ci51 ~mod* 0x25A1) - // cl51 = (ci73 ~mod* 0x25A1) ~mod+ (ci51 ~mod* 0x1925) - // ck1 = cl51 ~mod+ ((bq1 ~mod* 0x133E) ~mod+ (bq7 ~mod* 0xFFFFE333)) - // ck3 = cl73 ~mod+ ((bq3 ~mod* 0x1051) ~mod+ (bq5 ~mod* 0xFFFFADFD)) - // ck5 = cl51 ~mod+ ((bq3 ~mod* 0xFFFFADFD) ~mod+ (bq5 ~mod* 0xFFFFEFB0)) - // ck7 = cl73 ~mod+ ((bq1 ~mod* 0xFFFFE333) ~mod+ (bq7 ~mod* 0xFFFFECC1)) + // cl73 = (ci73 ~mod* 0xFFFF_E6DC) ~mod+ (ci51 ~mod* 0x0000_25A1) + // cl51 = (ci73 ~mod* 0x0000_25A1) ~mod+ (ci51 ~mod* 0x0000_1925) + // ck1 = cl51 ~mod+ ((bq1 ~mod* 0x0000_133E) ~mod+ (bq7 ~mod* 0xFFFF_E333)) + // ck3 = cl73 ~mod+ ((bq3 ~mod* 0x0000_1051) ~mod+ (bq5 ~mod* 0xFFFF_ADFD)) + // ck5 = cl51 ~mod+ ((bq3 ~mod* 0xFFFF_ADFD) ~mod+ (bq5 ~mod* 0xFFFF_EFB0)) + // ck7 = cl73 ~mod+ ((bq1 ~mod* 0xFFFF_E333) ~mod+ (bq7 ~mod* 0xFFFF_ECC1)) // Combine rows. @@ -575,15 +547,13 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q bq6 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x34]) ~mod* (this.quant_tables[args.q][0x34] as base.u32) // This code... - ca = (bq2 ~mod+ bq6) ~mod* 0x1151 - - cb2 = ca ~mod+ (bq2 ~mod* 0x187E) - cb6 = ca ~mod- (bq6 ~mod* 0x3B21) - + ca = (bq2 ~mod+ bq6) ~mod* 0x0000_1151 + cb2 = ca ~mod+ (bq2 ~mod* 0x0000_187E) + cb6 = ca ~mod- (bq6 ~mod* 0x0000_3B21) // ...is equivalent to this more-SIMD-like code. // - // cb2 = (bq2 ~mod* 0x29CF) ~mod+ (bq6 ~mod* 0x1151) - // cb6 = (bq2 ~mod* 0x1151) ~mod+ (bq6 ~mod* 0xFFFFD630) + // cb2 = (bq2 ~mod* 0x0000_29CF) ~mod+ (bq6 ~mod* 0x0000_1151) + // cb6 = (bq2 ~mod* 0x0000_1151) ~mod+ (bq6 ~mod* 0xFFFF_D630) bq0 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x04]) ~mod* (this.quant_tables[args.q][0x04] as base.u32) bq4 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x24]) ~mod* (this.quant_tables[args.q][0x24] as base.u32) @@ -609,34 +579,29 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q ci73 = bq7 ~mod+ bq3 // This code... - cj = (ci73 ~mod+ ci51) ~mod* 0x25A1 - - ck1 = bq1 ~mod* 0x300B - ck3 = bq3 ~mod* 0x6254 - ck5 = bq5 ~mod* 0x41B3 - ck7 = bq7 ~mod* 0x098E - + cj = (ci73 ~mod+ ci51) ~mod* 0x0000_25A1 + ck1 = bq1 ~mod* 0x0000_300B + ck3 = bq3 ~mod* 0x0000_6254 + ck5 = bq5 ~mod* 0x0000_41B3 + ck7 = bq7 ~mod* 0x0000_098E ci51 ~mod*= 0xFFFF_F384 ci53 ~mod*= 0xFFFF_ADFD ci71 ~mod*= 0xFFFF_E333 ci73 ~mod*= 0xFFFF_C13B - cl51 = ci51 ~mod+ cj cl73 = ci73 ~mod+ cj - ck1 ~mod+= ci71 ~mod+ cl51 ck3 ~mod+= ci53 ~mod+ cl73 ck5 ~mod+= ci53 ~mod+ cl51 ck7 ~mod+= ci71 ~mod+ cl73 - // ...is equivalent to this more-SIMD-like code. // - // cl73 = (ci73 ~mod* 0xFFFFE6DC) ~mod+ (ci51 ~mod* 0x25A1) - // cl51 = (ci73 ~mod* 0x25A1) ~mod+ (ci51 ~mod* 0x1925) - // ck1 = cl51 ~mod+ ((bq1 ~mod* 0x133E) ~mod+ (bq7 ~mod* 0xFFFFE333)) - // ck3 = cl73 ~mod+ ((bq3 ~mod* 0x1051) ~mod+ (bq5 ~mod* 0xFFFFADFD)) - // ck5 = cl51 ~mod+ ((bq3 ~mod* 0xFFFFADFD) ~mod+ (bq5 ~mod* 0xFFFFEFB0)) - // ck7 = cl73 ~mod+ ((bq1 ~mod* 0xFFFFE333) ~mod+ (bq7 ~mod* 0xFFFFECC1)) + // cl73 = (ci73 ~mod* 0xFFFF_E6DC) ~mod+ (ci51 ~mod* 0x0000_25A1) + // cl51 = (ci73 ~mod* 0x0000_25A1) ~mod+ (ci51 ~mod* 0x0000_1925) + // ck1 = cl51 ~mod+ ((bq1 ~mod* 0x0000_133E) ~mod+ (bq7 ~mod* 0xFFFF_E333)) + // ck3 = cl73 ~mod+ ((bq3 ~mod* 0x0000_1051) ~mod+ (bq5 ~mod* 0xFFFF_ADFD)) + // ck5 = cl51 ~mod+ ((bq3 ~mod* 0xFFFF_ADFD) ~mod+ (bq5 ~mod* 0xFFFF_EFB0)) + // ck7 = cl73 ~mod+ ((bq1 ~mod* 0xFFFF_E333) ~mod+ (bq7 ~mod* 0xFFFF_ECC1)) // Combine rows. @@ -680,15 +645,13 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q bq6 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x35]) ~mod* (this.quant_tables[args.q][0x35] as base.u32) // This code... - ca = (bq2 ~mod+ bq6) ~mod* 0x1151 - - cb2 = ca ~mod+ (bq2 ~mod* 0x187E) - cb6 = ca ~mod- (bq6 ~mod* 0x3B21) - + ca = (bq2 ~mod+ bq6) ~mod* 0x0000_1151 + cb2 = ca ~mod+ (bq2 ~mod* 0x0000_187E) + cb6 = ca ~mod- (bq6 ~mod* 0x0000_3B21) // ...is equivalent to this more-SIMD-like code. // - // cb2 = (bq2 ~mod* 0x29CF) ~mod+ (bq6 ~mod* 0x1151) - // cb6 = (bq2 ~mod* 0x1151) ~mod+ (bq6 ~mod* 0xFFFFD630) + // cb2 = (bq2 ~mod* 0x0000_29CF) ~mod+ (bq6 ~mod* 0x0000_1151) + // cb6 = (bq2 ~mod* 0x0000_1151) ~mod+ (bq6 ~mod* 0xFFFF_D630) bq0 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x05]) ~mod* (this.quant_tables[args.q][0x05] as base.u32) bq4 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x25]) ~mod* (this.quant_tables[args.q][0x25] as base.u32) @@ -714,34 +677,29 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q ci73 = bq7 ~mod+ bq3 // This code... - cj = (ci73 ~mod+ ci51) ~mod* 0x25A1 - - ck1 = bq1 ~mod* 0x300B - ck3 = bq3 ~mod* 0x6254 - ck5 = bq5 ~mod* 0x41B3 - ck7 = bq7 ~mod* 0x098E - + cj = (ci73 ~mod+ ci51) ~mod* 0x0000_25A1 + ck1 = bq1 ~mod* 0x0000_300B + ck3 = bq3 ~mod* 0x0000_6254 + ck5 = bq5 ~mod* 0x0000_41B3 + ck7 = bq7 ~mod* 0x0000_098E ci51 ~mod*= 0xFFFF_F384 ci53 ~mod*= 0xFFFF_ADFD ci71 ~mod*= 0xFFFF_E333 ci73 ~mod*= 0xFFFF_C13B - cl51 = ci51 ~mod+ cj cl73 = ci73 ~mod+ cj - ck1 ~mod+= ci71 ~mod+ cl51 ck3 ~mod+= ci53 ~mod+ cl73 ck5 ~mod+= ci53 ~mod+ cl51 ck7 ~mod+= ci71 ~mod+ cl73 - // ...is equivalent to this more-SIMD-like code. // - // cl73 = (ci73 ~mod* 0xFFFFE6DC) ~mod+ (ci51 ~mod* 0x25A1) - // cl51 = (ci73 ~mod* 0x25A1) ~mod+ (ci51 ~mod* 0x1925) - // ck1 = cl51 ~mod+ ((bq1 ~mod* 0x133E) ~mod+ (bq7 ~mod* 0xFFFFE333)) - // ck3 = cl73 ~mod+ ((bq3 ~mod* 0x1051) ~mod+ (bq5 ~mod* 0xFFFFADFD)) - // ck5 = cl51 ~mod+ ((bq3 ~mod* 0xFFFFADFD) ~mod+ (bq5 ~mod* 0xFFFFEFB0)) - // ck7 = cl73 ~mod+ ((bq1 ~mod* 0xFFFFE333) ~mod+ (bq7 ~mod* 0xFFFFECC1)) + // cl73 = (ci73 ~mod* 0xFFFF_E6DC) ~mod+ (ci51 ~mod* 0x0000_25A1) + // cl51 = (ci73 ~mod* 0x0000_25A1) ~mod+ (ci51 ~mod* 0x0000_1925) + // ck1 = cl51 ~mod+ ((bq1 ~mod* 0x0000_133E) ~mod+ (bq7 ~mod* 0xFFFF_E333)) + // ck3 = cl73 ~mod+ ((bq3 ~mod* 0x0000_1051) ~mod+ (bq5 ~mod* 0xFFFF_ADFD)) + // ck5 = cl51 ~mod+ ((bq3 ~mod* 0xFFFF_ADFD) ~mod+ (bq5 ~mod* 0xFFFF_EFB0)) + // ck7 = cl73 ~mod+ ((bq1 ~mod* 0xFFFF_E333) ~mod+ (bq7 ~mod* 0xFFFF_ECC1)) // Combine rows. @@ -785,15 +743,13 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q bq6 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x36]) ~mod* (this.quant_tables[args.q][0x36] as base.u32) // This code... - ca = (bq2 ~mod+ bq6) ~mod* 0x1151 - - cb2 = ca ~mod+ (bq2 ~mod* 0x187E) - cb6 = ca ~mod- (bq6 ~mod* 0x3B21) - + ca = (bq2 ~mod+ bq6) ~mod* 0x0000_1151 + cb2 = ca ~mod+ (bq2 ~mod* 0x0000_187E) + cb6 = ca ~mod- (bq6 ~mod* 0x0000_3B21) // ...is equivalent to this more-SIMD-like code. // - // cb2 = (bq2 ~mod* 0x29CF) ~mod+ (bq6 ~mod* 0x1151) - // cb6 = (bq2 ~mod* 0x1151) ~mod+ (bq6 ~mod* 0xFFFFD630) + // cb2 = (bq2 ~mod* 0x0000_29CF) ~mod+ (bq6 ~mod* 0x0000_1151) + // cb6 = (bq2 ~mod* 0x0000_1151) ~mod+ (bq6 ~mod* 0xFFFF_D630) bq0 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x06]) ~mod* (this.quant_tables[args.q][0x06] as base.u32) bq4 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x26]) ~mod* (this.quant_tables[args.q][0x26] as base.u32) @@ -819,34 +775,29 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q ci73 = bq7 ~mod+ bq3 // This code... - cj = (ci73 ~mod+ ci51) ~mod* 0x25A1 - - ck1 = bq1 ~mod* 0x300B - ck3 = bq3 ~mod* 0x6254 - ck5 = bq5 ~mod* 0x41B3 - ck7 = bq7 ~mod* 0x098E - + cj = (ci73 ~mod+ ci51) ~mod* 0x0000_25A1 + ck1 = bq1 ~mod* 0x0000_300B + ck3 = bq3 ~mod* 0x0000_6254 + ck5 = bq5 ~mod* 0x0000_41B3 + ck7 = bq7 ~mod* 0x0000_098E ci51 ~mod*= 0xFFFF_F384 ci53 ~mod*= 0xFFFF_ADFD ci71 ~mod*= 0xFFFF_E333 ci73 ~mod*= 0xFFFF_C13B - cl51 = ci51 ~mod+ cj cl73 = ci73 ~mod+ cj - ck1 ~mod+= ci71 ~mod+ cl51 ck3 ~mod+= ci53 ~mod+ cl73 ck5 ~mod+= ci53 ~mod+ cl51 ck7 ~mod+= ci71 ~mod+ cl73 - // ...is equivalent to this more-SIMD-like code. // - // cl73 = (ci73 ~mod* 0xFFFFE6DC) ~mod+ (ci51 ~mod* 0x25A1) - // cl51 = (ci73 ~mod* 0x25A1) ~mod+ (ci51 ~mod* 0x1925) - // ck1 = cl51 ~mod+ ((bq1 ~mod* 0x133E) ~mod+ (bq7 ~mod* 0xFFFFE333)) - // ck3 = cl73 ~mod+ ((bq3 ~mod* 0x1051) ~mod+ (bq5 ~mod* 0xFFFFADFD)) - // ck5 = cl51 ~mod+ ((bq3 ~mod* 0xFFFFADFD) ~mod+ (bq5 ~mod* 0xFFFFEFB0)) - // ck7 = cl73 ~mod+ ((bq1 ~mod* 0xFFFFE333) ~mod+ (bq7 ~mod* 0xFFFFECC1)) + // cl73 = (ci73 ~mod* 0xFFFF_E6DC) ~mod+ (ci51 ~mod* 0x0000_25A1) + // cl51 = (ci73 ~mod* 0x0000_25A1) ~mod+ (ci51 ~mod* 0x0000_1925) + // ck1 = cl51 ~mod+ ((bq1 ~mod* 0x0000_133E) ~mod+ (bq7 ~mod* 0xFFFF_E333)) + // ck3 = cl73 ~mod+ ((bq3 ~mod* 0x0000_1051) ~mod+ (bq5 ~mod* 0xFFFF_ADFD)) + // ck5 = cl51 ~mod+ ((bq3 ~mod* 0xFFFF_ADFD) ~mod+ (bq5 ~mod* 0xFFFF_EFB0)) + // ck7 = cl73 ~mod+ ((bq1 ~mod* 0xFFFF_E333) ~mod+ (bq7 ~mod* 0xFFFF_ECC1)) // Combine rows. @@ -890,15 +841,13 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q bq6 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x37]) ~mod* (this.quant_tables[args.q][0x37] as base.u32) // This code... - ca = (bq2 ~mod+ bq6) ~mod* 0x1151 - - cb2 = ca ~mod+ (bq2 ~mod* 0x187E) - cb6 = ca ~mod- (bq6 ~mod* 0x3B21) - + ca = (bq2 ~mod+ bq6) ~mod* 0x0000_1151 + cb2 = ca ~mod+ (bq2 ~mod* 0x0000_187E) + cb6 = ca ~mod- (bq6 ~mod* 0x0000_3B21) // ...is equivalent to this more-SIMD-like code. // - // cb2 = (bq2 ~mod* 0x29CF) ~mod+ (bq6 ~mod* 0x1151) - // cb6 = (bq2 ~mod* 0x1151) ~mod+ (bq6 ~mod* 0xFFFFD630) + // cb2 = (bq2 ~mod* 0x0000_29CF) ~mod+ (bq6 ~mod* 0x0000_1151) + // cb6 = (bq2 ~mod* 0x0000_1151) ~mod+ (bq6 ~mod* 0xFFFF_D630) bq0 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x07]) ~mod* (this.quant_tables[args.q][0x07] as base.u32) bq4 = this.util.sign_extend_convert_u16_u32(a: this.mcu_blocks[0][0x27]) ~mod* (this.quant_tables[args.q][0x27] as base.u32) @@ -924,34 +873,29 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q ci73 = bq7 ~mod+ bq3 // This code... - cj = (ci73 ~mod+ ci51) ~mod* 0x25A1 - - ck1 = bq1 ~mod* 0x300B - ck3 = bq3 ~mod* 0x6254 - ck5 = bq5 ~mod* 0x41B3 - ck7 = bq7 ~mod* 0x098E - + cj = (ci73 ~mod+ ci51) ~mod* 0x0000_25A1 + ck1 = bq1 ~mod* 0x0000_300B + ck3 = bq3 ~mod* 0x0000_6254 + ck5 = bq5 ~mod* 0x0000_41B3 + ck7 = bq7 ~mod* 0x0000_098E ci51 ~mod*= 0xFFFF_F384 ci53 ~mod*= 0xFFFF_ADFD ci71 ~mod*= 0xFFFF_E333 ci73 ~mod*= 0xFFFF_C13B - cl51 = ci51 ~mod+ cj cl73 = ci73 ~mod+ cj - ck1 ~mod+= ci71 ~mod+ cl51 ck3 ~mod+= ci53 ~mod+ cl73 ck5 ~mod+= ci53 ~mod+ cl51 ck7 ~mod+= ci71 ~mod+ cl73 - // ...is equivalent to this more-SIMD-like code. // - // cl73 = (ci73 ~mod* 0xFFFFE6DC) ~mod+ (ci51 ~mod* 0x25A1) - // cl51 = (ci73 ~mod* 0x25A1) ~mod+ (ci51 ~mod* 0x1925) - // ck1 = cl51 ~mod+ ((bq1 ~mod* 0x133E) ~mod+ (bq7 ~mod* 0xFFFFE333)) - // ck3 = cl73 ~mod+ ((bq3 ~mod* 0x1051) ~mod+ (bq5 ~mod* 0xFFFFADFD)) - // ck5 = cl51 ~mod+ ((bq3 ~mod* 0xFFFFADFD) ~mod+ (bq5 ~mod* 0xFFFFEFB0)) - // ck7 = cl73 ~mod+ ((bq1 ~mod* 0xFFFFE333) ~mod+ (bq7 ~mod* 0xFFFFECC1)) + // cl73 = (ci73 ~mod* 0xFFFF_E6DC) ~mod+ (ci51 ~mod* 0x0000_25A1) + // cl51 = (ci73 ~mod* 0x0000_25A1) ~mod+ (ci51 ~mod* 0x0000_1925) + // ck1 = cl51 ~mod+ ((bq1 ~mod* 0x0000_133E) ~mod+ (bq7 ~mod* 0xFFFF_E333)) + // ck3 = cl73 ~mod+ ((bq3 ~mod* 0x0000_1051) ~mod+ (bq5 ~mod* 0xFFFF_ADFD)) + // ck5 = cl51 ~mod+ ((bq3 ~mod* 0xFFFF_ADFD) ~mod+ (bq5 ~mod* 0xFFFF_EFB0)) + // ck7 = cl73 ~mod+ ((bq1 ~mod* 0xFFFF_E333) ~mod+ (bq7 ~mod* 0xFFFF_ECC1)) // Combine rows. @@ -1000,15 +944,13 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q in6 = intermediate[0x06] // This code... - ra = (in2 ~mod+ in6) ~mod* 0x1151 - - rb2 = ra ~mod+ (in2 ~mod* 0x187E) - rb6 = ra ~mod- (in6 ~mod* 0x3B21) - + ra = (in2 ~mod+ in6) ~mod* 0x0000_1151 + rb2 = ra ~mod+ (in2 ~mod* 0x0000_187E) + rb6 = ra ~mod- (in6 ~mod* 0x0000_3B21) // ...is equivalent to this more-SIMD-like code. // - // rb2 = (in2 ~mod* 0x29CF) ~mod+ (in6 ~mod* 0x1151) - // rb6 = (in2 ~mod* 0x1151) ~mod+ (in6 ~mod* 0xFFFFD630) + // rb2 = (in2 ~mod* 0x0000_29CF) ~mod+ (in6 ~mod* 0x0000_1151) + // rb6 = (in2 ~mod* 0x0000_1151) ~mod+ (in6 ~mod* 0xFFFF_D630) in0 = intermediate[0x00] in4 = intermediate[0x04] @@ -1034,34 +976,29 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q ri73 = in7 ~mod+ in3 // This code... - rj = (ri73 ~mod+ ri51) ~mod* 0x25A1 - - rk1 = in1 ~mod* 0x300B - rk3 = in3 ~mod* 0x6254 - rk5 = in5 ~mod* 0x41B3 - rk7 = in7 ~mod* 0x098E - + rj = (ri73 ~mod+ ri51) ~mod* 0x0000_25A1 + rk1 = in1 ~mod* 0x0000_300B + rk3 = in3 ~mod* 0x0000_6254 + rk5 = in5 ~mod* 0x0000_41B3 + rk7 = in7 ~mod* 0x0000_098E ri51 ~mod*= 0xFFFF_F384 ri53 ~mod*= 0xFFFF_ADFD ri71 ~mod*= 0xFFFF_E333 ri73 ~mod*= 0xFFFF_C13B - rl51 = ri51 ~mod+ rj rl73 = ri73 ~mod+ rj - rk1 ~mod+= ri71 ~mod+ rl51 rk3 ~mod+= ri53 ~mod+ rl73 rk5 ~mod+= ri53 ~mod+ rl51 rk7 ~mod+= ri71 ~mod+ rl73 - // ...is equivalent to this more-SIMD-like code. // - // rl73 = (ri73 ~mod* 0xFFFFE6DC) ~mod+ (ri51 ~mod* 0x25A1) - // rl51 = (ri73 ~mod* 0x25A1) ~mod+ (ri51 ~mod* 0x1925) - // rk1 = rl51 ~mod+ ((in1 ~mod* 0x133E) ~mod+ (in7 ~mod* 0xFFFFE333)) - // rk3 = rl73 ~mod+ ((in3 ~mod* 0x1051) ~mod+ (in5 ~mod* 0xFFFFADFD)) - // rk5 = rl51 ~mod+ ((in3 ~mod* 0xFFFFADFD) ~mod+ (in5 ~mod* 0xFFFFEFB0)) - // rk7 = rl73 ~mod+ ((in1 ~mod* 0xFFFFE333) ~mod+ (in7 ~mod* 0xFFFFECC1)) + // rl73 = (ri73 ~mod* 0xFFFF_E6DC) ~mod+ (ri51 ~mod* 0x0000_25A1) + // rl51 = (ri73 ~mod* 0x0000_25A1) ~mod+ (ri51 ~mod* 0x0000_1925) + // rk1 = rl51 ~mod+ ((in1 ~mod* 0x0000_133E) ~mod+ (in7 ~mod* 0xFFFF_E333)) + // rk3 = rl73 ~mod+ ((in3 ~mod* 0x0000_1051) ~mod+ (in5 ~mod* 0xFFFF_ADFD)) + // rk5 = rl51 ~mod+ ((in3 ~mod* 0xFFFF_ADFD) ~mod+ (in5 ~mod* 0xFFFF_EFB0)) + // rk7 = rl73 ~mod+ ((in1 ~mod* 0xFFFF_E333) ~mod+ (in7 ~mod* 0xFFFF_ECC1)) // Combine columns. @@ -1117,15 +1054,13 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q in6 = intermediate[0x0E] // This code... - ra = (in2 ~mod+ in6) ~mod* 0x1151 - - rb2 = ra ~mod+ (in2 ~mod* 0x187E) - rb6 = ra ~mod- (in6 ~mod* 0x3B21) - + ra = (in2 ~mod+ in6) ~mod* 0x0000_1151 + rb2 = ra ~mod+ (in2 ~mod* 0x0000_187E) + rb6 = ra ~mod- (in6 ~mod* 0x0000_3B21) // ...is equivalent to this more-SIMD-like code. // - // rb2 = (in2 ~mod* 0x29CF) ~mod+ (in6 ~mod* 0x1151) - // rb6 = (in2 ~mod* 0x1151) ~mod+ (in6 ~mod* 0xFFFFD630) + // rb2 = (in2 ~mod* 0x0000_29CF) ~mod+ (in6 ~mod* 0x0000_1151) + // rb6 = (in2 ~mod* 0x0000_1151) ~mod+ (in6 ~mod* 0xFFFF_D630) in0 = intermediate[0x08] in4 = intermediate[0x0C] @@ -1151,34 +1086,29 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q ri73 = in7 ~mod+ in3 // This code... - rj = (ri73 ~mod+ ri51) ~mod* 0x25A1 - - rk1 = in1 ~mod* 0x300B - rk3 = in3 ~mod* 0x6254 - rk5 = in5 ~mod* 0x41B3 - rk7 = in7 ~mod* 0x098E - + rj = (ri73 ~mod+ ri51) ~mod* 0x0000_25A1 + rk1 = in1 ~mod* 0x0000_300B + rk3 = in3 ~mod* 0x0000_6254 + rk5 = in5 ~mod* 0x0000_41B3 + rk7 = in7 ~mod* 0x0000_098E ri51 ~mod*= 0xFFFF_F384 ri53 ~mod*= 0xFFFF_ADFD ri71 ~mod*= 0xFFFF_E333 ri73 ~mod*= 0xFFFF_C13B - rl51 = ri51 ~mod+ rj rl73 = ri73 ~mod+ rj - rk1 ~mod+= ri71 ~mod+ rl51 rk3 ~mod+= ri53 ~mod+ rl73 rk5 ~mod+= ri53 ~mod+ rl51 rk7 ~mod+= ri71 ~mod+ rl73 - // ...is equivalent to this more-SIMD-like code. // - // rl73 = (ri73 ~mod* 0xFFFFE6DC) ~mod+ (ri51 ~mod* 0x25A1) - // rl51 = (ri73 ~mod* 0x25A1) ~mod+ (ri51 ~mod* 0x1925) - // rk1 = rl51 ~mod+ ((in1 ~mod* 0x133E) ~mod+ (in7 ~mod* 0xFFFFE333)) - // rk3 = rl73 ~mod+ ((in3 ~mod* 0x1051) ~mod+ (in5 ~mod* 0xFFFFADFD)) - // rk5 = rl51 ~mod+ ((in3 ~mod* 0xFFFFADFD) ~mod+ (in5 ~mod* 0xFFFFEFB0)) - // rk7 = rl73 ~mod+ ((in1 ~mod* 0xFFFFE333) ~mod+ (in7 ~mod* 0xFFFFECC1)) + // rl73 = (ri73 ~mod* 0xFFFF_E6DC) ~mod+ (ri51 ~mod* 0x0000_25A1) + // rl51 = (ri73 ~mod* 0x0000_25A1) ~mod+ (ri51 ~mod* 0x0000_1925) + // rk1 = rl51 ~mod+ ((in1 ~mod* 0x0000_133E) ~mod+ (in7 ~mod* 0xFFFF_E333)) + // rk3 = rl73 ~mod+ ((in3 ~mod* 0x0000_1051) ~mod+ (in5 ~mod* 0xFFFF_ADFD)) + // rk5 = rl51 ~mod+ ((in3 ~mod* 0xFFFF_ADFD) ~mod+ (in5 ~mod* 0xFFFF_EFB0)) + // rk7 = rl73 ~mod+ ((in1 ~mod* 0xFFFF_E333) ~mod+ (in7 ~mod* 0xFFFF_ECC1)) // Combine columns. @@ -1234,15 +1164,13 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q in6 = intermediate[0x16] // This code... - ra = (in2 ~mod+ in6) ~mod* 0x1151 - - rb2 = ra ~mod+ (in2 ~mod* 0x187E) - rb6 = ra ~mod- (in6 ~mod* 0x3B21) - + ra = (in2 ~mod+ in6) ~mod* 0x0000_1151 + rb2 = ra ~mod+ (in2 ~mod* 0x0000_187E) + rb6 = ra ~mod- (in6 ~mod* 0x0000_3B21) // ...is equivalent to this more-SIMD-like code. // - // rb2 = (in2 ~mod* 0x29CF) ~mod+ (in6 ~mod* 0x1151) - // rb6 = (in2 ~mod* 0x1151) ~mod+ (in6 ~mod* 0xFFFFD630) + // rb2 = (in2 ~mod* 0x0000_29CF) ~mod+ (in6 ~mod* 0x0000_1151) + // rb6 = (in2 ~mod* 0x0000_1151) ~mod+ (in6 ~mod* 0xFFFF_D630) in0 = intermediate[0x10] in4 = intermediate[0x14] @@ -1268,34 +1196,29 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q ri73 = in7 ~mod+ in3 // This code... - rj = (ri73 ~mod+ ri51) ~mod* 0x25A1 - - rk1 = in1 ~mod* 0x300B - rk3 = in3 ~mod* 0x6254 - rk5 = in5 ~mod* 0x41B3 - rk7 = in7 ~mod* 0x098E - + rj = (ri73 ~mod+ ri51) ~mod* 0x0000_25A1 + rk1 = in1 ~mod* 0x0000_300B + rk3 = in3 ~mod* 0x0000_6254 + rk5 = in5 ~mod* 0x0000_41B3 + rk7 = in7 ~mod* 0x0000_098E ri51 ~mod*= 0xFFFF_F384 ri53 ~mod*= 0xFFFF_ADFD ri71 ~mod*= 0xFFFF_E333 ri73 ~mod*= 0xFFFF_C13B - rl51 = ri51 ~mod+ rj rl73 = ri73 ~mod+ rj - rk1 ~mod+= ri71 ~mod+ rl51 rk3 ~mod+= ri53 ~mod+ rl73 rk5 ~mod+= ri53 ~mod+ rl51 rk7 ~mod+= ri71 ~mod+ rl73 - // ...is equivalent to this more-SIMD-like code. // - // rl73 = (ri73 ~mod* 0xFFFFE6DC) ~mod+ (ri51 ~mod* 0x25A1) - // rl51 = (ri73 ~mod* 0x25A1) ~mod+ (ri51 ~mod* 0x1925) - // rk1 = rl51 ~mod+ ((in1 ~mod* 0x133E) ~mod+ (in7 ~mod* 0xFFFFE333)) - // rk3 = rl73 ~mod+ ((in3 ~mod* 0x1051) ~mod+ (in5 ~mod* 0xFFFFADFD)) - // rk5 = rl51 ~mod+ ((in3 ~mod* 0xFFFFADFD) ~mod+ (in5 ~mod* 0xFFFFEFB0)) - // rk7 = rl73 ~mod+ ((in1 ~mod* 0xFFFFE333) ~mod+ (in7 ~mod* 0xFFFFECC1)) + // rl73 = (ri73 ~mod* 0xFFFF_E6DC) ~mod+ (ri51 ~mod* 0x0000_25A1) + // rl51 = (ri73 ~mod* 0x0000_25A1) ~mod+ (ri51 ~mod* 0x0000_1925) + // rk1 = rl51 ~mod+ ((in1 ~mod* 0x0000_133E) ~mod+ (in7 ~mod* 0xFFFF_E333)) + // rk3 = rl73 ~mod+ ((in3 ~mod* 0x0000_1051) ~mod+ (in5 ~mod* 0xFFFF_ADFD)) + // rk5 = rl51 ~mod+ ((in3 ~mod* 0xFFFF_ADFD) ~mod+ (in5 ~mod* 0xFFFF_EFB0)) + // rk7 = rl73 ~mod+ ((in1 ~mod* 0xFFFF_E333) ~mod+ (in7 ~mod* 0xFFFF_ECC1)) // Combine columns. @@ -1351,15 +1274,13 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q in6 = intermediate[0x1E] // This code... - ra = (in2 ~mod+ in6) ~mod* 0x1151 - - rb2 = ra ~mod+ (in2 ~mod* 0x187E) - rb6 = ra ~mod- (in6 ~mod* 0x3B21) - + ra = (in2 ~mod+ in6) ~mod* 0x0000_1151 + rb2 = ra ~mod+ (in2 ~mod* 0x0000_187E) + rb6 = ra ~mod- (in6 ~mod* 0x0000_3B21) // ...is equivalent to this more-SIMD-like code. // - // rb2 = (in2 ~mod* 0x29CF) ~mod+ (in6 ~mod* 0x1151) - // rb6 = (in2 ~mod* 0x1151) ~mod+ (in6 ~mod* 0xFFFFD630) + // rb2 = (in2 ~mod* 0x0000_29CF) ~mod+ (in6 ~mod* 0x0000_1151) + // rb6 = (in2 ~mod* 0x0000_1151) ~mod+ (in6 ~mod* 0xFFFF_D630) in0 = intermediate[0x18] in4 = intermediate[0x1C] @@ -1385,34 +1306,29 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q ri73 = in7 ~mod+ in3 // This code... - rj = (ri73 ~mod+ ri51) ~mod* 0x25A1 - - rk1 = in1 ~mod* 0x300B - rk3 = in3 ~mod* 0x6254 - rk5 = in5 ~mod* 0x41B3 - rk7 = in7 ~mod* 0x098E - + rj = (ri73 ~mod+ ri51) ~mod* 0x0000_25A1 + rk1 = in1 ~mod* 0x0000_300B + rk3 = in3 ~mod* 0x0000_6254 + rk5 = in5 ~mod* 0x0000_41B3 + rk7 = in7 ~mod* 0x0000_098E ri51 ~mod*= 0xFFFF_F384 ri53 ~mod*= 0xFFFF_ADFD ri71 ~mod*= 0xFFFF_E333 ri73 ~mod*= 0xFFFF_C13B - rl51 = ri51 ~mod+ rj rl73 = ri73 ~mod+ rj - rk1 ~mod+= ri71 ~mod+ rl51 rk3 ~mod+= ri53 ~mod+ rl73 rk5 ~mod+= ri53 ~mod+ rl51 rk7 ~mod+= ri71 ~mod+ rl73 - // ...is equivalent to this more-SIMD-like code. // - // rl73 = (ri73 ~mod* 0xFFFFE6DC) ~mod+ (ri51 ~mod* 0x25A1) - // rl51 = (ri73 ~mod* 0x25A1) ~mod+ (ri51 ~mod* 0x1925) - // rk1 = rl51 ~mod+ ((in1 ~mod* 0x133E) ~mod+ (in7 ~mod* 0xFFFFE333)) - // rk3 = rl73 ~mod+ ((in3 ~mod* 0x1051) ~mod+ (in5 ~mod* 0xFFFFADFD)) - // rk5 = rl51 ~mod+ ((in3 ~mod* 0xFFFFADFD) ~mod+ (in5 ~mod* 0xFFFFEFB0)) - // rk7 = rl73 ~mod+ ((in1 ~mod* 0xFFFFE333) ~mod+ (in7 ~mod* 0xFFFFECC1)) + // rl73 = (ri73 ~mod* 0xFFFF_E6DC) ~mod+ (ri51 ~mod* 0x0000_25A1) + // rl51 = (ri73 ~mod* 0x0000_25A1) ~mod+ (ri51 ~mod* 0x0000_1925) + // rk1 = rl51 ~mod+ ((in1 ~mod* 0x0000_133E) ~mod+ (in7 ~mod* 0xFFFF_E333)) + // rk3 = rl73 ~mod+ ((in3 ~mod* 0x0000_1051) ~mod+ (in5 ~mod* 0xFFFF_ADFD)) + // rk5 = rl51 ~mod+ ((in3 ~mod* 0xFFFF_ADFD) ~mod+ (in5 ~mod* 0xFFFF_EFB0)) + // rk7 = rl73 ~mod+ ((in1 ~mod* 0xFFFF_E333) ~mod+ (in7 ~mod* 0xFFFF_ECC1)) // Combine columns. @@ -1468,15 +1384,13 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q in6 = intermediate[0x26] // This code... - ra = (in2 ~mod+ in6) ~mod* 0x1151 - - rb2 = ra ~mod+ (in2 ~mod* 0x187E) - rb6 = ra ~mod- (in6 ~mod* 0x3B21) - + ra = (in2 ~mod+ in6) ~mod* 0x0000_1151 + rb2 = ra ~mod+ (in2 ~mod* 0x0000_187E) + rb6 = ra ~mod- (in6 ~mod* 0x0000_3B21) // ...is equivalent to this more-SIMD-like code. // - // rb2 = (in2 ~mod* 0x29CF) ~mod+ (in6 ~mod* 0x1151) - // rb6 = (in2 ~mod* 0x1151) ~mod+ (in6 ~mod* 0xFFFFD630) + // rb2 = (in2 ~mod* 0x0000_29CF) ~mod+ (in6 ~mod* 0x0000_1151) + // rb6 = (in2 ~mod* 0x0000_1151) ~mod+ (in6 ~mod* 0xFFFF_D630) in0 = intermediate[0x20] in4 = intermediate[0x24] @@ -1502,34 +1416,29 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q ri73 = in7 ~mod+ in3 // This code... - rj = (ri73 ~mod+ ri51) ~mod* 0x25A1 - - rk1 = in1 ~mod* 0x300B - rk3 = in3 ~mod* 0x6254 - rk5 = in5 ~mod* 0x41B3 - rk7 = in7 ~mod* 0x098E - + rj = (ri73 ~mod+ ri51) ~mod* 0x0000_25A1 + rk1 = in1 ~mod* 0x0000_300B + rk3 = in3 ~mod* 0x0000_6254 + rk5 = in5 ~mod* 0x0000_41B3 + rk7 = in7 ~mod* 0x0000_098E ri51 ~mod*= 0xFFFF_F384 ri53 ~mod*= 0xFFFF_ADFD ri71 ~mod*= 0xFFFF_E333 ri73 ~mod*= 0xFFFF_C13B - rl51 = ri51 ~mod+ rj rl73 = ri73 ~mod+ rj - rk1 ~mod+= ri71 ~mod+ rl51 rk3 ~mod+= ri53 ~mod+ rl73 rk5 ~mod+= ri53 ~mod+ rl51 rk7 ~mod+= ri71 ~mod+ rl73 - // ...is equivalent to this more-SIMD-like code. // - // rl73 = (ri73 ~mod* 0xFFFFE6DC) ~mod+ (ri51 ~mod* 0x25A1) - // rl51 = (ri73 ~mod* 0x25A1) ~mod+ (ri51 ~mod* 0x1925) - // rk1 = rl51 ~mod+ ((in1 ~mod* 0x133E) ~mod+ (in7 ~mod* 0xFFFFE333)) - // rk3 = rl73 ~mod+ ((in3 ~mod* 0x1051) ~mod+ (in5 ~mod* 0xFFFFADFD)) - // rk5 = rl51 ~mod+ ((in3 ~mod* 0xFFFFADFD) ~mod+ (in5 ~mod* 0xFFFFEFB0)) - // rk7 = rl73 ~mod+ ((in1 ~mod* 0xFFFFE333) ~mod+ (in7 ~mod* 0xFFFFECC1)) + // rl73 = (ri73 ~mod* 0xFFFF_E6DC) ~mod+ (ri51 ~mod* 0x0000_25A1) + // rl51 = (ri73 ~mod* 0x0000_25A1) ~mod+ (ri51 ~mod* 0x0000_1925) + // rk1 = rl51 ~mod+ ((in1 ~mod* 0x0000_133E) ~mod+ (in7 ~mod* 0xFFFF_E333)) + // rk3 = rl73 ~mod+ ((in3 ~mod* 0x0000_1051) ~mod+ (in5 ~mod* 0xFFFF_ADFD)) + // rk5 = rl51 ~mod+ ((in3 ~mod* 0xFFFF_ADFD) ~mod+ (in5 ~mod* 0xFFFF_EFB0)) + // rk7 = rl73 ~mod+ ((in1 ~mod* 0xFFFF_E333) ~mod+ (in7 ~mod* 0xFFFF_ECC1)) // Combine columns. @@ -1585,15 +1494,13 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q in6 = intermediate[0x2E] // This code... - ra = (in2 ~mod+ in6) ~mod* 0x1151 - - rb2 = ra ~mod+ (in2 ~mod* 0x187E) - rb6 = ra ~mod- (in6 ~mod* 0x3B21) - + ra = (in2 ~mod+ in6) ~mod* 0x0000_1151 + rb2 = ra ~mod+ (in2 ~mod* 0x0000_187E) + rb6 = ra ~mod- (in6 ~mod* 0x0000_3B21) // ...is equivalent to this more-SIMD-like code. // - // rb2 = (in2 ~mod* 0x29CF) ~mod+ (in6 ~mod* 0x1151) - // rb6 = (in2 ~mod* 0x1151) ~mod+ (in6 ~mod* 0xFFFFD630) + // rb2 = (in2 ~mod* 0x0000_29CF) ~mod+ (in6 ~mod* 0x0000_1151) + // rb6 = (in2 ~mod* 0x0000_1151) ~mod+ (in6 ~mod* 0xFFFF_D630) in0 = intermediate[0x28] in4 = intermediate[0x2C] @@ -1619,34 +1526,29 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q ri73 = in7 ~mod+ in3 // This code... - rj = (ri73 ~mod+ ri51) ~mod* 0x25A1 - - rk1 = in1 ~mod* 0x300B - rk3 = in3 ~mod* 0x6254 - rk5 = in5 ~mod* 0x41B3 - rk7 = in7 ~mod* 0x098E - + rj = (ri73 ~mod+ ri51) ~mod* 0x0000_25A1 + rk1 = in1 ~mod* 0x0000_300B + rk3 = in3 ~mod* 0x0000_6254 + rk5 = in5 ~mod* 0x0000_41B3 + rk7 = in7 ~mod* 0x0000_098E ri51 ~mod*= 0xFFFF_F384 ri53 ~mod*= 0xFFFF_ADFD ri71 ~mod*= 0xFFFF_E333 ri73 ~mod*= 0xFFFF_C13B - rl51 = ri51 ~mod+ rj rl73 = ri73 ~mod+ rj - rk1 ~mod+= ri71 ~mod+ rl51 rk3 ~mod+= ri53 ~mod+ rl73 rk5 ~mod+= ri53 ~mod+ rl51 rk7 ~mod+= ri71 ~mod+ rl73 - // ...is equivalent to this more-SIMD-like code. // - // rl73 = (ri73 ~mod* 0xFFFFE6DC) ~mod+ (ri51 ~mod* 0x25A1) - // rl51 = (ri73 ~mod* 0x25A1) ~mod+ (ri51 ~mod* 0x1925) - // rk1 = rl51 ~mod+ ((in1 ~mod* 0x133E) ~mod+ (in7 ~mod* 0xFFFFE333)) - // rk3 = rl73 ~mod+ ((in3 ~mod* 0x1051) ~mod+ (in5 ~mod* 0xFFFFADFD)) - // rk5 = rl51 ~mod+ ((in3 ~mod* 0xFFFFADFD) ~mod+ (in5 ~mod* 0xFFFFEFB0)) - // rk7 = rl73 ~mod+ ((in1 ~mod* 0xFFFFE333) ~mod+ (in7 ~mod* 0xFFFFECC1)) + // rl73 = (ri73 ~mod* 0xFFFF_E6DC) ~mod+ (ri51 ~mod* 0x0000_25A1) + // rl51 = (ri73 ~mod* 0x0000_25A1) ~mod+ (ri51 ~mod* 0x0000_1925) + // rk1 = rl51 ~mod+ ((in1 ~mod* 0x0000_133E) ~mod+ (in7 ~mod* 0xFFFF_E333)) + // rk3 = rl73 ~mod+ ((in3 ~mod* 0x0000_1051) ~mod+ (in5 ~mod* 0xFFFF_ADFD)) + // rk5 = rl51 ~mod+ ((in3 ~mod* 0xFFFF_ADFD) ~mod+ (in5 ~mod* 0xFFFF_EFB0)) + // rk7 = rl73 ~mod+ ((in1 ~mod* 0xFFFF_E333) ~mod+ (in7 ~mod* 0xFFFF_ECC1)) // Combine columns. @@ -1702,15 +1604,13 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q in6 = intermediate[0x36] // This code... - ra = (in2 ~mod+ in6) ~mod* 0x1151 - - rb2 = ra ~mod+ (in2 ~mod* 0x187E) - rb6 = ra ~mod- (in6 ~mod* 0x3B21) - + ra = (in2 ~mod+ in6) ~mod* 0x0000_1151 + rb2 = ra ~mod+ (in2 ~mod* 0x0000_187E) + rb6 = ra ~mod- (in6 ~mod* 0x0000_3B21) // ...is equivalent to this more-SIMD-like code. // - // rb2 = (in2 ~mod* 0x29CF) ~mod+ (in6 ~mod* 0x1151) - // rb6 = (in2 ~mod* 0x1151) ~mod+ (in6 ~mod* 0xFFFFD630) + // rb2 = (in2 ~mod* 0x0000_29CF) ~mod+ (in6 ~mod* 0x0000_1151) + // rb6 = (in2 ~mod* 0x0000_1151) ~mod+ (in6 ~mod* 0xFFFF_D630) in0 = intermediate[0x30] in4 = intermediate[0x34] @@ -1736,34 +1636,29 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q ri73 = in7 ~mod+ in3 // This code... - rj = (ri73 ~mod+ ri51) ~mod* 0x25A1 - - rk1 = in1 ~mod* 0x300B - rk3 = in3 ~mod* 0x6254 - rk5 = in5 ~mod* 0x41B3 - rk7 = in7 ~mod* 0x098E - + rj = (ri73 ~mod+ ri51) ~mod* 0x0000_25A1 + rk1 = in1 ~mod* 0x0000_300B + rk3 = in3 ~mod* 0x0000_6254 + rk5 = in5 ~mod* 0x0000_41B3 + rk7 = in7 ~mod* 0x0000_098E ri51 ~mod*= 0xFFFF_F384 ri53 ~mod*= 0xFFFF_ADFD ri71 ~mod*= 0xFFFF_E333 ri73 ~mod*= 0xFFFF_C13B - rl51 = ri51 ~mod+ rj rl73 = ri73 ~mod+ rj - rk1 ~mod+= ri71 ~mod+ rl51 rk3 ~mod+= ri53 ~mod+ rl73 rk5 ~mod+= ri53 ~mod+ rl51 rk7 ~mod+= ri71 ~mod+ rl73 - // ...is equivalent to this more-SIMD-like code. // - // rl73 = (ri73 ~mod* 0xFFFFE6DC) ~mod+ (ri51 ~mod* 0x25A1) - // rl51 = (ri73 ~mod* 0x25A1) ~mod+ (ri51 ~mod* 0x1925) - // rk1 = rl51 ~mod+ ((in1 ~mod* 0x133E) ~mod+ (in7 ~mod* 0xFFFFE333)) - // rk3 = rl73 ~mod+ ((in3 ~mod* 0x1051) ~mod+ (in5 ~mod* 0xFFFFADFD)) - // rk5 = rl51 ~mod+ ((in3 ~mod* 0xFFFFADFD) ~mod+ (in5 ~mod* 0xFFFFEFB0)) - // rk7 = rl73 ~mod+ ((in1 ~mod* 0xFFFFE333) ~mod+ (in7 ~mod* 0xFFFFECC1)) + // rl73 = (ri73 ~mod* 0xFFFF_E6DC) ~mod+ (ri51 ~mod* 0x0000_25A1) + // rl51 = (ri73 ~mod* 0x0000_25A1) ~mod+ (ri51 ~mod* 0x0000_1925) + // rk1 = rl51 ~mod+ ((in1 ~mod* 0x0000_133E) ~mod+ (in7 ~mod* 0xFFFF_E333)) + // rk3 = rl73 ~mod+ ((in3 ~mod* 0x0000_1051) ~mod+ (in5 ~mod* 0xFFFF_ADFD)) + // rk5 = rl51 ~mod+ ((in3 ~mod* 0xFFFF_ADFD) ~mod+ (in5 ~mod* 0xFFFF_EFB0)) + // rk7 = rl73 ~mod+ ((in1 ~mod* 0xFFFF_E333) ~mod+ (in7 ~mod* 0xFFFF_ECC1)) // Combine columns. @@ -1816,15 +1711,13 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q in6 = intermediate[0x3E] // This code... - ra = (in2 ~mod+ in6) ~mod* 0x1151 - - rb2 = ra ~mod+ (in2 ~mod* 0x187E) - rb6 = ra ~mod- (in6 ~mod* 0x3B21) - + ra = (in2 ~mod+ in6) ~mod* 0x0000_1151 + rb2 = ra ~mod+ (in2 ~mod* 0x0000_187E) + rb6 = ra ~mod- (in6 ~mod* 0x0000_3B21) // ...is equivalent to this more-SIMD-like code. // - // rb2 = (in2 ~mod* 0x29CF) ~mod+ (in6 ~mod* 0x1151) - // rb6 = (in2 ~mod* 0x1151) ~mod+ (in6 ~mod* 0xFFFFD630) + // rb2 = (in2 ~mod* 0x0000_29CF) ~mod+ (in6 ~mod* 0x0000_1151) + // rb6 = (in2 ~mod* 0x0000_1151) ~mod+ (in6 ~mod* 0xFFFF_D630) in0 = intermediate[0x38] in4 = intermediate[0x3C] @@ -1850,34 +1743,29 @@ pri func decoder.decode_idct!(dst_buffer: slice base.u8, dst_stride: base.u64, q ri73 = in7 ~mod+ in3 // This code... - rj = (ri73 ~mod+ ri51) ~mod* 0x25A1 - - rk1 = in1 ~mod* 0x300B - rk3 = in3 ~mod* 0x6254 - rk5 = in5 ~mod* 0x41B3 - rk7 = in7 ~mod* 0x098E - + rj = (ri73 ~mod+ ri51) ~mod* 0x0000_25A1 + rk1 = in1 ~mod* 0x0000_300B + rk3 = in3 ~mod* 0x0000_6254 + rk5 = in5 ~mod* 0x0000_41B3 + rk7 = in7 ~mod* 0x0000_098E ri51 ~mod*= 0xFFFF_F384 ri53 ~mod*= 0xFFFF_ADFD ri71 ~mod*= 0xFFFF_E333 ri73 ~mod*= 0xFFFF_C13B - rl51 = ri51 ~mod+ rj rl73 = ri73 ~mod+ rj - rk1 ~mod+= ri71 ~mod+ rl51 rk3 ~mod+= ri53 ~mod+ rl73 rk5 ~mod+= ri53 ~mod+ rl51 rk7 ~mod+= ri71 ~mod+ rl73 - // ...is equivalent to this more-SIMD-like code. // - // rl73 = (ri73 ~mod* 0xFFFFE6DC) ~mod+ (ri51 ~mod* 0x25A1) - // rl51 = (ri73 ~mod* 0x25A1) ~mod+ (ri51 ~mod* 0x1925) - // rk1 = rl51 ~mod+ ((in1 ~mod* 0x133E) ~mod+ (in7 ~mod* 0xFFFFE333)) - // rk3 = rl73 ~mod+ ((in3 ~mod* 0x1051) ~mod+ (in5 ~mod* 0xFFFFADFD)) - // rk5 = rl51 ~mod+ ((in3 ~mod* 0xFFFFADFD) ~mod+ (in5 ~mod* 0xFFFFEFB0)) - // rk7 = rl73 ~mod+ ((in1 ~mod* 0xFFFFE333) ~mod+ (in7 ~mod* 0xFFFFECC1)) + // rl73 = (ri73 ~mod* 0xFFFF_E6DC) ~mod+ (ri51 ~mod* 0x0000_25A1) + // rl51 = (ri73 ~mod* 0x0000_25A1) ~mod+ (ri51 ~mod* 0x0000_1925) + // rk1 = rl51 ~mod+ ((in1 ~mod* 0x0000_133E) ~mod+ (in7 ~mod* 0xFFFF_E333)) + // rk3 = rl73 ~mod+ ((in3 ~mod* 0x0000_1051) ~mod+ (in5 ~mod* 0xFFFF_ADFD)) + // rk5 = rl51 ~mod+ ((in3 ~mod* 0xFFFF_ADFD) ~mod+ (in5 ~mod* 0xFFFF_EFB0)) + // rk7 = rl73 ~mod+ ((in1 ~mod* 0xFFFF_E333) ~mod+ (in7 ~mod* 0xFFFF_ECC1)) // Combine columns.