SIMD-accelerated floating point quantize and convsamp routines for MIPS DSPr2

git-svn-id: svn://svn.code.sf.net/p/libjpeg-turbo/code/trunk@1058 632fc199-4ca6-4c93-a231-07263d6284db

4 files changed, 519 insertions, 0 deletions

diff --git a/simd/jsimd.h b/simd/jsimd.h
index ec32a8f..e5c5d44 100644
--- a/simd/jsimd.h
+++ b/simd/jsimd.h
@@ -663,6 +663,10 @@ EXTERN(void) jsimd_convsamp_float_sse2 JPP((JSAMPARRAY sample_data,
                                             JDIMENSION start_col,
                                             FAST_FLOAT * workspace));
 
+EXTERN(void) jsimd_convsamp_float_mips_dspr2 JPP((JSAMPARRAY sample_data,
+                                                  JDIMENSION start_col,
+                                                  FAST_FLOAT * workspace));
+
 /* SIMD Forward DCT */
 EXTERN(void) jsimd_fdct_islow_mmx JPP((DCTELEM * data));
 EXTERN(void) jsimd_fdct_ifast_mmx JPP((DCTELEM * data));
@@ -711,6 +715,10 @@ EXTERN(void) jsimd_quantize_float_sse2 JPP((JCOEFPTR coef_block,
                                             FAST_FLOAT * divisors,
                                             FAST_FLOAT * workspace));
 
+EXTERN(void) jsimd_quantize_float_mips_dspr2 JPP((JCOEFPTR coef_block,
+                                                  FAST_FLOAT * divisors,
+                                                  FAST_FLOAT * workspace));
+
 /* SIMD Reduced Inverse DCT */
 EXTERN(void) jsimd_idct_2x2_mmx JPP((void * dct_table,
                                      JCOEFPTR coef_block,
diff --git a/simd/jsimd_mips.c b/simd/jsimd_mips.c
index f3b5afe..d8d6b19 100644
--- a/simd/jsimd_mips.c
+++ b/simd/jsimd_mips.c
@@ -459,6 +459,23 @@ jsimd_can_convsamp (void)
 GLOBAL(int)
 jsimd_can_convsamp_float (void)
 {
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(JCOEF) != 2)
+    return 0;
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if (sizeof(ISLOW_MULT_TYPE) != 2)
+    return 0;
+
+  if ((simd_support & JSIMD_MIPS_DSPR2))
+    return 1;
+
   return 0;
 }
 
@@ -472,6 +489,8 @@ GLOBAL(void)
 jsimd_convsamp_float (JSAMPARRAY sample_data, JDIMENSION start_col,
                       FAST_FLOAT * workspace)
 {
+  if ((simd_support & JSIMD_MIPS_DSPR2))
+    jsimd_convsamp_float_mips_dspr2(sample_data, start_col, workspace);
 }
 
 GLOBAL(int)
@@ -555,6 +574,23 @@ jsimd_can_quantize (void)
 GLOBAL(int)
 jsimd_can_quantize_float (void)
 {
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (DCTSIZE != 8)
+    return 0;
+  if (sizeof(JCOEF) != 2)
+    return 0;
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if (sizeof(ISLOW_MULT_TYPE) != 2)
+    return 0;
+
+  if ((simd_support & JSIMD_MIPS_DSPR2))
+    return 1;
+
   return 0;
 }
 
@@ -570,6 +606,8 @@ GLOBAL(void)
 jsimd_quantize_float (JCOEFPTR coef_block, FAST_FLOAT * divisors,
                       FAST_FLOAT * workspace)
 {
+  if ((simd_support & JSIMD_MIPS_DSPR2))
+    jsimd_quantize_float_mips_dspr2(coef_block, divisors, workspace);
 }
 
 GLOBAL(int)
diff --git a/simd/jsimd_mips_dspr2.S b/simd/jsimd_mips_dspr2.S
index bfedae7..d478a6d 100644
--- a/simd/jsimd_mips_dspr2.S
+++ b/simd/jsimd_mips_dspr2.S
@@ -1666,6 +1666,86 @@ LEAF_MIPS_DSPR2(jsimd_quantize_mips_dspr2)
 END(jsimd_quantize_mips_dspr2)
 
 /*****************************************************************************/
+LEAF_MIPS_DSPR2(jsimd_quantize_float_mips_dspr2)
+/*
+ * a0     - coef_block
+ * a1     - divisors
+ * a2     - workspace
+ */
+
+    .set at
+
+    li         t1, 0x46800100     //integer representation 16384.5
+    mtc1       t1, f0
+    li         t0, 63
+0:
+    lwc1       f1, 0(a2)
+    lwc1       f5, 0(a1)
+    lwc1       f2, 4(a2)
+    lwc1       f6, 4(a1)
+    lwc1       f3, 8(a2)
+    lwc1       f7, 8(a1)
+    lwc1       f4, 12(a2)
+    lwc1       f8, 12(a1)
+    madd.s     f1, f0, f1, f5
+    madd.s     f2, f0, f2, f6
+    madd.s     f3, f0, f3, f7
+    madd.s     f4, f0, f4, f8
+    lwc1       f5, 16(a1)
+    lwc1       f6, 20(a1)
+    trunc.w.s  f1, f1
+    trunc.w.s  f2, f2
+    trunc.w.s  f3, f3
+    trunc.w.s  f4, f4
+    lwc1       f7, 24(a1)
+    lwc1       f8, 28(a1)
+    mfc1       t1, f1
+    mfc1       t2, f2
+    mfc1       t3, f3
+    mfc1       t4, f4
+    lwc1       f1, 16(a2)
+    lwc1       f2, 20(a2)
+    lwc1       f3, 24(a2)
+    lwc1       f4, 28(a2)
+    madd.s     f1, f0, f1, f5
+    madd.s     f2, f0, f2, f6
+    madd.s     f3, f0, f3, f7
+    madd.s     f4, f0, f4, f8
+    addiu      t1, t1, -16384
+    addiu      t2, t2, -16384
+    addiu      t3, t3, -16384
+    addiu      t4, t4, -16384
+    trunc.w.s  f1, f1
+    trunc.w.s  f2, f2
+    trunc.w.s  f3, f3
+    trunc.w.s  f4, f4
+    sh         t1, 0(a0)
+    sh         t2, 2(a0)
+    sh         t3, 4(a0)
+    sh         t4, 6(a0)
+    mfc1       t1, f1
+    mfc1       t2, f2
+    mfc1       t3, f3
+    mfc1       t4, f4
+    addiu      t0, t0, -8
+    addiu      a2, a2, 32
+    addiu      a1, a1, 32
+    addiu      t1, t1, -16384
+    addiu      t2, t2, -16384
+    addiu      t3, t3, -16384
+    addiu      t4, t4, -16384
+    sh         t1, 8(a0)
+    sh         t2, 10(a0)
+    sh         t3, 12(a0)
+    sh         t4, 14(a0)
+    bgez       t0, 0b
+     addiu     a0, a0, 16
+
+    j          ra
+     nop
+
+END(jsimd_quantize_float_mips_dspr2)
+/*****************************************************************************/
 LEAF_MIPS_DSPR2(jsimd_idct_2x2_mips_dspr2)
 /*
  * a0     - compptr->dct_table
@@ -2733,3 +2813,363 @@ LEAF_MIPS_DSPR2(jsimd_idct_12x12_pass2_mips_dspr2)
 END(jsimd_idct_12x12_pass2_mips_dspr2)
 
 /*****************************************************************************/
+LEAF_MIPS_DSPR2(jsimd_convsamp_float_mips_dspr2)
+/*
+ * a0     - sample_data
+ * a1     - start_col
+ * a2     - workspace
+ */
+
+    .set at
+
+    lw       t0, 0(a0)
+    addu     t0, t0, a1
+    lbu      t1, 0(t0)
+    lbu      t2, 1(t0)
+    lbu      t3, 2(t0)
+    lbu      t4, 3(t0)
+    lbu      t5, 4(t0)
+    lbu      t6, 5(t0)
+    lbu      t7, 6(t0)
+    lbu      t8, 7(t0)
+    addiu    t1, t1, -128
+    addiu    t2, t2, -128
+    addiu    t3, t3, -128
+    addiu    t4, t4, -128
+    addiu    t5, t5, -128
+    addiu    t6, t6, -128
+    addiu    t7, t7, -128
+    addiu    t8, t8, -128
+    mtc1     t1, f1
+    mtc1     t2, f2
+    mtc1     t3, f3
+    mtc1     t4, f4
+    mtc1     t5, f5
+    mtc1     t6, f6
+    mtc1     t7, f7
+    mtc1     t8, f8
+    cvt.s.w  f1, f1
+    cvt.s.w  f2, f2
+    cvt.s.w  f3, f3
+    cvt.s.w  f4, f4
+    cvt.s.w  f5, f5
+    cvt.s.w  f6, f6
+    cvt.s.w  f7, f7
+    cvt.s.w  f8, f8
+    lw       t0, 4(a0)
+    swc1     f1, 0(a2)
+    swc1     f2, 4(a2)
+    swc1     f3, 8(a2)
+    addu     t0, t0, a1
+    swc1     f4, 12(a2)
+    swc1     f5, 16(a2)
+    swc1     f6, 20(a2)
+    swc1     f7, 24(a2)
+    swc1     f8, 28(a2)
+    //elemr 1
+    lbu      t1, 0(t0)
+    lbu      t2, 1(t0)
+    lbu      t3, 2(t0)
+    lbu      t4, 3(t0)
+    lbu      t5, 4(t0)
+    lbu      t6, 5(t0)
+    lbu      t7, 6(t0)
+    lbu      t8, 7(t0)
+    addiu    t1, t1, -128
+    addiu    t2, t2, -128
+    addiu    t3, t3, -128
+    addiu    t4, t4, -128
+    addiu    t5, t5, -128
+    addiu    t6, t6, -128
+    addiu    t7, t7, -128
+    addiu    t8, t8, -128
+    mtc1     t1, f1
+    mtc1     t2, f2
+    mtc1     t3, f3
+    mtc1     t4, f4
+    mtc1     t5, f5
+    mtc1     t6, f6
+    mtc1     t7, f7
+    mtc1     t8, f8
+    cvt.s.w  f1, f1
+    cvt.s.w  f2, f2
+    cvt.s.w  f3, f3
+    cvt.s.w  f4, f4
+    cvt.s.w  f5, f5
+    cvt.s.w  f6, f6
+    cvt.s.w  f7, f7
+    cvt.s.w  f8, f8
+    lw       t0, 8(a0)
+    swc1     f1, 32(a2)
+    swc1     f2, 36(a2)
+    swc1     f3, 40(a2)
+    addu     t0, t0, a1
+    swc1     f4, 44(a2)
+    swc1     f5, 48(a2)
+    swc1     f6, 52(a2)
+    swc1     f7, 56(a2)
+    swc1     f8, 60(a2)
+    //elemr 2
+    lbu      t1, 0(t0)
+    lbu      t2, 1(t0)
+    lbu      t3, 2(t0)
+    lbu      t4, 3(t0)
+    lbu      t5, 4(t0)
+    lbu      t6, 5(t0)
+    lbu      t7, 6(t0)
+    lbu      t8, 7(t0)
+    addiu    t1, t1, -128
+    addiu    t2, t2, -128
+    addiu    t3, t3, -128
+    addiu    t4, t4, -128
+    addiu    t5, t5, -128
+    addiu    t6, t6, -128
+    addiu    t7, t7, -128
+    addiu    t8, t8, -128
+    mtc1     t1, f1
+    mtc1     t2, f2
+    mtc1     t3, f3
+    mtc1     t4, f4
+    mtc1     t5, f5
+    mtc1     t6, f6
+    mtc1     t7, f7
+    mtc1     t8, f8
+    cvt.s.w  f1, f1
+    cvt.s.w  f2, f2
+    cvt.s.w  f3, f3
+    cvt.s.w  f4, f4
+    cvt.s.w  f5, f5
+    cvt.s.w  f6, f6
+    cvt.s.w  f7, f7
+    cvt.s.w  f8, f8
+    lw       t0, 12(a0)
+    swc1     f1, 64(a2)
+    swc1     f2, 68(a2)
+    swc1     f3, 72(a2)
+    addu     t0, t0, a1
+    swc1     f4, 76(a2)
+    swc1     f5, 80(a2)
+    swc1     f6, 84(a2)
+    swc1     f7, 88(a2)
+    swc1     f8, 92(a2)
+    //elemr 3
+    lbu      t1, 0(t0)
+    lbu      t2, 1(t0)
+    lbu      t3, 2(t0)
+    lbu      t4, 3(t0)
+    lbu      t5, 4(t0)
+    lbu      t6, 5(t0)
+    lbu      t7, 6(t0)
+    lbu      t8, 7(t0)
+    addiu    t1, t1, -128
+    addiu    t2, t2, -128
+    addiu    t3, t3, -128
+    addiu    t4, t4, -128
+    addiu    t5, t5, -128
+    addiu    t6, t6, -128
+    addiu    t7, t7, -128
+    addiu    t8, t8, -128
+    mtc1     t1, f1
+    mtc1     t2, f2
+    mtc1     t3, f3
+    mtc1     t4, f4
+    mtc1     t5, f5
+    mtc1     t6, f6
+    mtc1     t7, f7
+    mtc1     t8, f8
+    cvt.s.w  f1, f1
+    cvt.s.w  f2, f2
+    cvt.s.w  f3, f3
+    cvt.s.w  f4, f4
+    cvt.s.w  f5, f5
+    cvt.s.w  f6, f6
+    cvt.s.w  f7, f7
+    cvt.s.w  f8, f8
+    lw       t0, 16(a0)
+    swc1     f1, 96(a2)
+    swc1     f2, 100(a2)
+    swc1     f3, 104(a2)
+    addu     t0, t0, a1
+    swc1     f4, 108(a2)
+    swc1     f5, 112(a2)
+    swc1     f6, 116(a2)
+    swc1     f7, 120(a2)
+    swc1     f8, 124(a2)
+    //elemr 4
+    lbu      t1, 0(t0)
+    lbu      t2, 1(t0)
+    lbu      t3, 2(t0)
+    lbu      t4, 3(t0)
+    lbu      t5, 4(t0)
+    lbu      t6, 5(t0)
+    lbu      t7, 6(t0)
+    lbu      t8, 7(t0)
+    addiu    t1, t1, -128
+    addiu    t2, t2, -128
+    addiu    t3, t3, -128
+    addiu    t4, t4, -128
+    addiu    t5, t5, -128
+    addiu    t6, t6, -128
+    addiu    t7, t7, -128
+    addiu    t8, t8, -128
+    mtc1     t1, f1
+    mtc1     t2, f2
+    mtc1     t3, f3
+    mtc1     t4, f4
+    mtc1     t5, f5
+    mtc1     t6, f6
+    mtc1     t7, f7
+    mtc1     t8, f8
+    cvt.s.w  f1, f1
+    cvt.s.w  f2, f2
+    cvt.s.w  f3, f3
+    cvt.s.w  f4, f4
+    cvt.s.w  f5, f5
+    cvt.s.w  f6, f6
+    cvt.s.w  f7, f7
+    cvt.s.w  f8, f8
+    lw       t0, 20(a0)
+    swc1     f1, 128(a2)
+    swc1     f2, 132(a2)
+    swc1     f3, 136(a2)
+    addu     t0, t0, a1
+    swc1     f4, 140(a2)
+    swc1     f5, 144(a2)
+    swc1     f6, 148(a2)
+    swc1     f7, 152(a2)
+    swc1     f8, 156(a2)
+    //elemr 5
+    lbu      t1, 0(t0)
+    lbu      t2, 1(t0)
+    lbu      t3, 2(t0)
+    lbu      t4, 3(t0)
+    lbu      t5, 4(t0)
+    lbu      t6, 5(t0)
+    lbu      t7, 6(t0)
+    lbu      t8, 7(t0)
+    addiu    t1, t1, -128
+    addiu    t2, t2, -128
+    addiu    t3, t3, -128
+    addiu    t4, t4, -128
+    addiu    t5, t5, -128
+    addiu    t6, t6, -128
+    addiu    t7, t7, -128
+    addiu    t8, t8, -128
+    mtc1     t1, f1
+    mtc1     t2, f2
+    mtc1     t3, f3
+    mtc1     t4, f4
+    mtc1     t5, f5
+    mtc1     t6, f6
+    mtc1     t7, f7
+    mtc1     t8, f8
+    cvt.s.w  f1, f1
+    cvt.s.w  f2, f2
+    cvt.s.w  f3, f3
+    cvt.s.w  f4, f4
+    cvt.s.w  f5, f5
+    cvt.s.w  f6, f6
+    cvt.s.w  f7, f7
+    cvt.s.w  f8, f8
+    lw       t0, 24(a0)
+    swc1     f1, 160(a2)
+    swc1     f2, 164(a2)
+    swc1     f3, 168(a2)
+    addu     t0, t0, a1
+    swc1     f4, 172(a2)
+    swc1     f5, 176(a2)
+    swc1     f6, 180(a2)
+    swc1     f7, 184(a2)
+    swc1     f8, 188(a2)
+    //elemr 6
+    lbu      t1, 0(t0)
+    lbu      t2, 1(t0)
+    lbu      t3, 2(t0)
+    lbu      t4, 3(t0)
+    lbu      t5, 4(t0)
+    lbu      t6, 5(t0)
+    lbu      t7, 6(t0)
+    lbu      t8, 7(t0)
+    addiu    t1, t1, -128
+    addiu    t2, t2, -128
+    addiu    t3, t3, -128
+    addiu    t4, t4, -128
+    addiu    t5, t5, -128
+    addiu    t6, t6, -128
+    addiu    t7, t7, -128
+    addiu    t8, t8, -128
+    mtc1     t1, f1
+    mtc1     t2, f2
+    mtc1     t3, f3
+    mtc1     t4, f4
+    mtc1     t5, f5
+    mtc1     t6, f6
+    mtc1     t7, f7
+    mtc1     t8, f8
+    cvt.s.w  f1, f1
+    cvt.s.w  f2, f2
+    cvt.s.w  f3, f3
+    cvt.s.w  f4, f4
+    cvt.s.w  f5, f5
+    cvt.s.w  f6, f6
+    cvt.s.w  f7, f7
+    cvt.s.w  f8, f8
+    lw       t0, 28(a0)
+    swc1     f1, 192(a2)
+    swc1     f2, 196(a2)
+    swc1     f3, 200(a2)
+    addu     t0, t0, a1
+    swc1     f4, 204(a2)
+    swc1     f5, 208(a2)
+    swc1     f6, 212(a2)
+    swc1     f7, 216(a2)
+    swc1     f8, 220(a2)
+    //elemr 7
+    lbu      t1, 0(t0)
+    lbu      t2, 1(t0)
+    lbu      t3, 2(t0)
+    lbu      t4, 3(t0)
+    lbu      t5, 4(t0)
+    lbu      t6, 5(t0)
+    lbu      t7, 6(t0)
+    lbu      t8, 7(t0)
+    addiu    t1, t1, -128
+    addiu    t2, t2, -128
+    addiu    t3, t3, -128
+    addiu    t4, t4, -128
+    addiu    t5, t5, -128
+    addiu    t6, t6, -128
+    addiu    t7, t7, -128
+    addiu    t8, t8, -128
+    mtc1     t1, f1
+    mtc1     t2, f2
+    mtc1     t3, f3
+    mtc1     t4, f4
+    mtc1     t5, f5
+    mtc1     t6, f6
+    mtc1     t7, f7
+    mtc1     t8, f8
+    cvt.s.w  f1, f1
+    cvt.s.w  f2, f2
+    cvt.s.w  f3, f3
+    cvt.s.w  f4, f4
+    cvt.s.w  f5, f5
+    cvt.s.w  f6, f6
+    cvt.s.w  f7, f7
+    cvt.s.w  f8, f8
+    swc1     f1, 224(a2)
+    swc1     f2, 228(a2)
+    swc1     f3, 232(a2)
+    swc1     f4, 236(a2)
+    swc1     f5, 240(a2)
+    swc1     f6, 244(a2)
+    swc1     f7, 248(a2)
+    swc1     f8, 252(a2)
+
+    j        ra
+     nop
+
+END(jsimd_convsamp_float_mips_dspr2)
+
+/*****************************************************************************/
+
diff --git a/simd/jsimd_mips_dspr2_asm.h b/simd/jsimd_mips_dspr2_asm.h
index 53cf2bc..50ec31b 100644
--- a/simd/jsimd_mips_dspr2_asm.h
+++ b/simd/jsimd_mips_dspr2_asm.h
@@ -56,6 +56,39 @@
 #define s8   $30
 #define ra   $31
 
+#define f0   $f0
+#define f1   $f1
+#define f2   $f2
+#define f3   $f3
+#define f4   $f4
+#define f5   $f5
+#define f6   $f6
+#define f7   $f7
+#define f8   $f8
+#define f9   $f9
+#define f10  $f10
+#define f11  $f11
+#define f12  $f12
+#define f13  $f13
+#define f14  $f14
+#define f15  $f15
+#define f16  $f16
+#define f17  $f17
+#define f18  $f18
+#define f19  $f19
+#define f20  $f20
+#define f21  $f21
+#define f22  $f22
+#define f23  $f23
+#define f24  $f24
+#define f25  $f25
+#define f26  $f26
+#define f27  $f27
+#define f28  $f28
+#define f29  $f29
+#define f30  $f30
+#define f31  $f31
+
 /*
  * LEAF_MIPS32R2 - declare leaf routine for MIPS32r2
  */