From f042d662e2cec4315ddaae1ee536f593139f703d Mon Sep 17 00:00:00 2001 From: Brian Paul Date: Thu, 3 Jul 2008 12:56:33 -0600 Subject: gallium: increase TGSI interpreter's number of temp registers to 64 Also, clean up the definitions of the misc/extra temp regs. A few new assertions too. --- src/gallium/auxiliary/tgsi/exec/tgsi_exec.c | 4 ++- src/gallium/auxiliary/tgsi/exec/tgsi_exec.h | 48 +++++++++++++++-------------- 2 files changed, 28 insertions(+), 24 deletions(-) (limited to 'src/gallium/auxiliary') diff --git a/src/gallium/auxiliary/tgsi/exec/tgsi_exec.c b/src/gallium/auxiliary/tgsi/exec/tgsi_exec.c index 9649396c9b..46949661af 100644 --- a/src/gallium/auxiliary/tgsi/exec/tgsi_exec.c +++ b/src/gallium/auxiliary/tgsi/exec/tgsi_exec.c @@ -262,7 +262,7 @@ tgsi_exec_machine_init( uint i; mach->Temps = (struct tgsi_exec_vector *) tgsi_align_128bit( mach->_Temps); - mach->Addrs = &mach->Temps[TGSI_EXEC_NUM_TEMPS]; + mach->Addrs = &mach->Temps[TGSI_EXEC_TEMP_ADDR]; /* Setup constants. */ for( i = 0; i < 4; i++ ) { @@ -941,6 +941,7 @@ fetch_src_file_channel( break; case TGSI_FILE_TEMPORARY: + assert(index->i[0] < TGSI_EXEC_NUM_TEMPS); chan->u[0] = mach->Temps[index->i[0]].xyzw[swizzle].u[0]; chan->u[1] = mach->Temps[index->i[1]].xyzw[swizzle].u[1]; chan->u[2] = mach->Temps[index->i[2]].xyzw[swizzle].u[2]; @@ -1127,6 +1128,7 @@ store_dest( break; case TGSI_FILE_TEMPORARY: + assert(reg->DstRegister.Index < TGSI_EXEC_NUM_TEMPS); dst = &mach->Temps[reg->DstRegister.Index].xyzw[chan_index]; break; diff --git a/src/gallium/auxiliary/tgsi/exec/tgsi_exec.h b/src/gallium/auxiliary/tgsi/exec/tgsi_exec.h index c89dfcb167..18abdd9ac0 100644 --- a/src/gallium/auxiliary/tgsi/exec/tgsi_exec.h +++ b/src/gallium/auxiliary/tgsi/exec/tgsi_exec.h @@ -99,53 +99,57 @@ struct tgsi_exec_labels unsigned count; }; + +#define TGSI_EXEC_NUM_TEMPS 64 +#define TGSI_EXEC_NUM_TEMP_EXTRAS 6 +#define TGSI_EXEC_NUM_IMMEDIATES 256 + /* * Locations of various utility registers (_I = Index, _C = Channel) */ -#define TGSI_EXEC_TEMP_00000000_I 32 +#define TGSI_EXEC_TEMP_00000000_I (TGSI_EXEC_NUM_TEMPS + 0) #define TGSI_EXEC_TEMP_00000000_C 0 -#define TGSI_EXEC_TEMP_7FFFFFFF_I 32 +#define TGSI_EXEC_TEMP_7FFFFFFF_I (TGSI_EXEC_NUM_TEMPS + 0) #define TGSI_EXEC_TEMP_7FFFFFFF_C 1 -#define TGSI_EXEC_TEMP_80000000_I 32 +#define TGSI_EXEC_TEMP_80000000_I (TGSI_EXEC_NUM_TEMPS + 0) #define TGSI_EXEC_TEMP_80000000_C 2 -#define TGSI_EXEC_TEMP_FFFFFFFF_I 32 +#define TGSI_EXEC_TEMP_FFFFFFFF_I (TGSI_EXEC_NUM_TEMPS + 0) #define TGSI_EXEC_TEMP_FFFFFFFF_C 3 -#define TGSI_EXEC_TEMP_ONE_I 33 +#define TGSI_EXEC_TEMP_ONE_I (TGSI_EXEC_NUM_TEMPS + 1) #define TGSI_EXEC_TEMP_ONE_C 0 -#define TGSI_EXEC_TEMP_TWO_I 33 +#define TGSI_EXEC_TEMP_TWO_I (TGSI_EXEC_NUM_TEMPS + 1) #define TGSI_EXEC_TEMP_TWO_C 1 -#define TGSI_EXEC_TEMP_128_I 33 +#define TGSI_EXEC_TEMP_128_I (TGSI_EXEC_NUM_TEMPS + 1) #define TGSI_EXEC_TEMP_128_C 2 -#define TGSI_EXEC_TEMP_MINUS_128_I 33 +#define TGSI_EXEC_TEMP_MINUS_128_I (TGSI_EXEC_NUM_TEMPS + 1) #define TGSI_EXEC_TEMP_MINUS_128_C 3 -#define TGSI_EXEC_TEMP_KILMASK_I 34 +#define TGSI_EXEC_TEMP_KILMASK_I (TGSI_EXEC_NUM_TEMPS + 2) #define TGSI_EXEC_TEMP_KILMASK_C 0 -#define TGSI_EXEC_TEMP_OUTPUT_I 34 +#define TGSI_EXEC_TEMP_OUTPUT_I (TGSI_EXEC_NUM_TEMPS + 2) #define TGSI_EXEC_TEMP_OUTPUT_C 1 -#define TGSI_EXEC_TEMP_PRIMITIVE_I 34 +#define TGSI_EXEC_TEMP_PRIMITIVE_I (TGSI_EXEC_NUM_TEMPS + 2) #define TGSI_EXEC_TEMP_PRIMITIVE_C 2 -#define TGSI_EXEC_TEMP_THREE_I 34 +#define TGSI_EXEC_TEMP_THREE_I (TGSI_EXEC_NUM_TEMPS + 2) #define TGSI_EXEC_TEMP_THREE_C 3 -#define TGSI_EXEC_TEMP_HALF_I 35 +#define TGSI_EXEC_TEMP_HALF_I (TGSI_EXEC_NUM_TEMPS + 3) #define TGSI_EXEC_TEMP_HALF_C 0 -#define TGSI_EXEC_TEMP_R0 36 +#define TGSI_EXEC_TEMP_R0 (TGSI_EXEC_NUM_TEMPS + 4) + +#define TGSI_EXEC_TEMP_ADDR (TGSI_EXEC_NUM_TEMPS + 5) -#define TGSI_EXEC_NUM_TEMPS (32 + 5) -#define TGSI_EXEC_NUM_ADDRS 1 -#define TGSI_EXEC_NUM_IMMEDIATES 256 #define TGSI_EXEC_MAX_COND_NESTING 20 #define TGSI_EXEC_MAX_LOOP_NESTING 20 @@ -156,13 +160,11 @@ struct tgsi_exec_labels */ struct tgsi_exec_machine { - /* - * 32 program temporaries - * 4 internal temporaries - * 1 address - * 1 temporary of padding to align to 16 bytes + /* Total = program temporaries + internal temporaries + * + 1 padding to align to 16 bytes */ - struct tgsi_exec_vector _Temps[TGSI_EXEC_NUM_TEMPS + TGSI_EXEC_NUM_ADDRS + 1]; + struct tgsi_exec_vector _Temps[TGSI_EXEC_NUM_TEMPS + + TGSI_EXEC_NUM_TEMP_EXTRAS + 1]; /* * This will point to _Temps after aligning to 16B boundary. -- cgit v1.2.3