ARM: use =const syntax instead of explicit literal pools

Signed-off-by: Mans Rullgard <mans@mansr.com>

1 files changed, 16 insertions, 40 deletions

diff --git a/libavcodec/arm/simple_idct_arm.S b/libavcodec/arm/simple_idct_arm.S
index c6540a1e40..8ba6c48b5a 100644
--- a/libavcodec/arm/simple_idct_arm.S
+++ b/libavcodec/arm/simple_idct_arm.S
@@ -25,8 +25,7 @@
 
 #include "libavutil/arm/asm.S"
 
-/* useful constants for the algorithm, they are save in __constant_ptr__ at */
-/* the end of the source code.*/
+/* useful constants for the algorithm */
 #define W1  22725
 #define W2  21407
 #define W3  19266
@@ -36,16 +35,6 @@
 #define W7  4520
 #define MASK_MSHW 0xFFFF0000
 
-/* offsets of the constants in the vector */
-#define offW1  0
-#define offW2  4
-#define offW3  8
-#define offW4  12
-#define offW5  16
-#define offW6  20
-#define offW7  24
-#define offMASK_MSHW 28
-
 #define ROW_SHIFT 11
 #define ROW_SHIFT2MSHW (16-11)
 #define COL_SHIFT 20
@@ -63,7 +52,6 @@ function ff_simple_idct_arm, export=1
         stmfd sp!, {r4-r11, r14} @ R14 is also called LR
         @@ at this point, R0=block, other registers are free.
         add r14, r0, #112        @ R14=&block[8*7], better start from the last row, and decrease the value until row=0, i.e. R12=block.
-        adr r12, __constant_ptr__ @ R12=__constant_ptr__, the vector containing the constants, probably not necessary to reserve a register for it
         @@ add 2 temporary variables in the stack: R0 and R14
         sub sp, sp, #8          @ allow 2 local variables
         str r0, [sp, #0]        @ save block in sp[0]
@@ -109,13 +97,13 @@ __b_evaluation:
         @@ MAC16(b1, -W7, row[3]);
         @@ MAC16(b2, -W1, row[3]);
         @@ MAC16(b3, -W5, row[3]);
-        ldr r8, [r12, #offW1]    @ R8=W1
+        ldr r8, =W1              @ R8=W1
         mov r2, r2, asr #16      @ R2=ROWr16[3]
         mul r0, r8, r7           @ R0=W1*ROWr16[1]=b0 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle)
-        ldr r9, [r12, #offW3]    @ R9=W3
-        ldr r10, [r12, #offW5]   @ R10=W5
+        ldr r9, =W3              @ R9=W3
+        ldr r10, =W5             @ R10=W5
         mul r1, r9, r7           @ R1=W3*ROWr16[1]=b1 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle)
-        ldr r11, [r12, #offW7]   @ R11=W7
+        ldr r11, =W7             @ R11=W7
         mul r5, r10, r7          @ R5=W5*ROWr16[1]=b2 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle)
         mul r7, r11, r7          @ R7=W7*ROWr16[1]=b3 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle)
         teq r2, #0               @ if null avoid muls
@@ -177,14 +165,14 @@ __a_evaluation:
         @@ a2 = a0 - W6 * row[2];
         @@ a3 = a0 - W2 * row[2];
         @@ a0 = a0 + W2 * row[2];
-        ldr r9, [r12, #offW4]    @ R9=W4
+        ldr r9, =W4              @ R9=W4
         mul r6, r9, r6           @ R6=W4*ROWr16[0]
-        ldr r10, [r12, #offW6]   @ R10=W6
+        ldr r10, =W6             @ R10=W6
         ldrsh r4, [r14, #4]      @ R4=ROWr16[2] (a3 not defined yet)
         add r6, r6, #ROW_SHIFTED_1 @ R6=W4*ROWr16[0] + 1<<(ROW_SHIFT-1) (a0)
 
         mul r11, r10, r4         @ R11=W6*ROWr16[2]
-        ldr r8, [r12, #offW2]    @ R8=W2
+        ldr r8, =W2              @ R8=W2
         sub r3, r6, r11          @ R3=a0-W6*ROWr16[2] (a2)
         @@ temp = ((uint32_t*)row)[2] | ((uint32_t*)row)[3];
         @@ if (temp != 0) {}
@@ -248,7 +236,7 @@ __end_a_evaluation:
         add r9, r2, r1           @ R9=a1+b1
         @@ put 2 16 bits half-words in a 32bits word
         @@ ROWr32[0]=ROWr16[0] | (ROWr16[1]<<16) (only Little Endian compliant then!!!)
-        ldr r10, [r12, #offMASK_MSHW] @ R10=0xFFFF0000
+        ldr r10, =MASK_MSHW      @ R10=0xFFFF0000
         and r9, r10, r9, lsl #ROW_SHIFT2MSHW @ R9=0xFFFF0000 & ((a1+b1)<<5)
         mvn r11, r10             @ R11= NOT R10= 0x0000FFFF
         and r8, r11, r8, asr #ROW_SHIFT @ R8=0x0000FFFF & ((a0+b0)>>11)
@@ -319,13 +307,13 @@ __b_evaluation2:
         @@ MAC16(b1, -W7, col[8x3]);
         @@ MAC16(b2, -W1, col[8x3]);
         @@ MAC16(b3, -W5, col[8x3]);
-        ldr r8, [r12, #offW1]    @ R8=W1
+        ldr r8, =W1              @ R8=W1
         ldrsh r7, [r14, #16]
         mul r0, r8, r7           @ R0=W1*ROWr16[1]=b0 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle)
-        ldr r9, [r12, #offW3]    @ R9=W3
-        ldr r10, [r12, #offW5]   @ R10=W5
+        ldr r9, =W3              @ R9=W3
+        ldr r10, =W5             @ R10=W5
         mul r1, r9, r7           @ R1=W3*ROWr16[1]=b1 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle)
-        ldr r11, [r12, #offW7]   @ R11=W7
+        ldr r11, =W7             @ R11=W7
         mul r5, r10, r7          @ R5=W5*ROWr16[1]=b2 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle)
         ldrsh r2, [r14, #48]
         mul r7, r11, r7          @ R7=W7*ROWr16[1]=b3 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle)
@@ -381,13 +369,13 @@ __a_evaluation2:
         @@ a3 = a0 - W2 * row[2];
         @@ a0 = a0 + W2 * row[2];
         ldrsh r6, [r14, #0]
-        ldr r9, [r12, #offW4]    @ R9=W4
+        ldr r9, =W4              @ R9=W4
         mul r6, r9, r6           @ R6=W4*ROWr16[0]
-        ldr r10, [r12, #offW6]   @ R10=W6
+        ldr r10, =W6             @ R10=W6
         ldrsh r4, [r14, #32]     @ R4=ROWr16[2] (a3 not defined yet)
         add r6, r6, #COL_SHIFTED_1 @ R6=W4*ROWr16[0] + 1<<(COL_SHIFT-1) (a0)
         mul r11, r10, r4         @ R11=W6*ROWr16[2]
-        ldr r8, [r12, #offW2]    @ R8=W2
+        ldr r8, =W2              @ R8=W2
         add r2, r6, r11          @ R2=a0+W6*ROWr16[2] (a1)
         sub r3, r6, r11          @ R3=a0-W6*ROWr16[2] (a2)
         mul r11, r8, r4          @ R11=W2*ROWr16[2]
@@ -489,15 +477,3 @@ __end_bef_a_evaluation:
         sub r4, r6, r11          @ R4=a0-W2*ROWr16[2] (a3)
         add r6, r6, r11          @ R6=a0+W2*ROWr16[2] (a0)
         bal __end_a_evaluation
-
-
-        .align
-__constant_ptr__:  @@ see #defines at the beginning of the source code for values.
-        .word   W1
-        .word   W2
-        .word   W3
-        .word   W4
-        .word   W5
-        .word   W6
-        .word   W7
-        .word   MASK_MSHW