diff options
| author | Ian Lance Taylor <ian@gcc.gnu.org> | 1996-06-04 18:12:36 +0000 |
|---|---|---|
| committer | Ian Lance Taylor <ian@gcc.gnu.org> | 1996-06-04 18:12:36 +0000 |
| commit | 446b03afbb0bc351c36a92c103d68e83f0f57937 (patch) | |
| tree | 6f221822b4f2a76accdf86b3ff60aa57f975e7ea /gcc | |
| parent | 24b867d8bed5b6b49c0d371637e1021abcc08595 (diff) | |
| download | gcc-446b03afbb0bc351c36a92c103d68e83f0f57937.zip gcc-446b03afbb0bc351c36a92c103d68e83f0f57937.tar.gz gcc-446b03afbb0bc351c36a92c103d68e83f0f57937.tar.bz2 | |
Align code at the start loops and after unconditional branches
From-SVN: r12154
Diffstat (limited to 'gcc')
| -rw-r--r-- | gcc/config/sh/sh.h | 46 |
1 files changed, 41 insertions, 5 deletions
diff --git a/gcc/config/sh/sh.h b/gcc/config/sh/sh.h index 4135585..1c7714b 100644 --- a/gcc/config/sh/sh.h +++ b/gcc/config/sh/sh.h @@ -219,9 +219,6 @@ do { \ /* Allocation boundary (in *bits*) for the code of a function. 32 bit alignment is faster, because instructions are always fetched as a pair from a longword boundary. */ -/* ??? Perhaps also define ASM_OUTPUT_ALIGN_CODE and/or ASM_OUTPUT_LOOP_ALIGN - so as to align jump targets and/or loops to 4 byte boundaries when not - optimizing for space? */ #define FUNCTION_BOUNDARY (TARGET_SMALLCODE ? 16 : 32) /* Alignment of field after `int : 0' in a structure. */ @@ -1437,6 +1434,17 @@ do { char dstr[30]; \ #define ASM_OUTPUT_BYTE(STREAM, VALUE) \ fprintf (STREAM, "\t.byte\t%d\n", VALUE) \ +/* Align loops and labels after unconditional branches to get faster + code. */ + +#define ASM_OUTPUT_LOOP_ALIGN(FILE) \ + if (! TARGET_SMALLCODE) \ + ASM_OUTPUT_ALIGN ((FILE), 2) + +#define ASM_OUTPUT_ALIGN_CODE(FILE) \ + if (! TARGET_SMALLCODE) \ + ASM_OUTPUT_ALIGN ((FILE), (TARGET_SH3 || TARGET_SH3E) ? 4 : 2) + /* This is how to output an assembler line that says to advance the location counter by SIZE bytes. */ @@ -1538,7 +1546,11 @@ extern int pragma_interrupt; #define MOVE_RATIO (TARGET_SMALLCODE ? 2 : 16) /* Instructions with unfilled delay slots take up an extra two bytes for - the nop in the delay slot. */ + the nop in the delay slot. Instructions at the start of loops, or + after unconditional branches, may take up extra room when they are + aligned. ??? We would get more accurate results if we did instruction + alignment based on the value of INSN_CURRENT_ADDRESS; the approach used + here is too conservative. */ #define ADJUST_INSN_LENGTH(X, LENGTH) \ if (((GET_CODE (X) == INSN \ @@ -1550,7 +1562,31 @@ extern int pragma_interrupt; && GET_CODE (PATTERN (X)) != ADDR_DIFF_VEC \ && GET_CODE (PATTERN (X)) != ADDR_VEC)) \ && get_attr_needs_delay_slot (X) == NEEDS_DELAY_SLOT_YES) \ - LENGTH += 2; + LENGTH += 2; \ + if (! TARGET_SMALLCODE) \ + { \ + rtx aip; \ + for (aip = PREV_INSN (X); aip; aip = PREV_INSN (aip)) \ + { \ + if (GET_CODE (aip) == BARRIER) \ + { \ + if (TARGET_SH3 || TARGET_SH3E) \ + LENGTH += 14; \ + else \ + LENGTH += 2; \ + break; \ + } \ + else if ((GET_CODE (aip) == NOTE \ + && NOTE_LINE_NUMBER (aip) == NOTE_INSN_LOOP_BEG)) \ + { \ + LENGTH += 2; \ + break; \ + } \ + else if (GET_CODE (aip) != NOTE \ + && GET_CODE (aip) != CODE_LABEL) \ + break; \ + } \ + } /* Enable a bug fix for the shorten_branches pass. */ #define SHORTEN_WITH_ADJUST_INSN_LENGTH |