diff options
| author | Wilson Fung <[email protected]> | 2012-11-12 21:56:33 -0800 |
|---|---|---|
| committer | Andrew Boktor <[email protected]> | 2014-08-14 13:49:21 -0700 |
| commit | 6e0aab40706497ef4e3046d5614b0061f538f300 (patch) | |
| tree | fc8dd9f49b471544ac8c5856fa18adfccd32f931 /src | |
| parent | 117dfa890501388dec301f461f0cca2562d6a941 (diff) | |
Changing the QuadroFX5800 config to use compute capability 1.3 (no idea why it was not...). Adding sign-extension mode for cvt.s16.s32 that writes to a .u32 register. Adding stub parsing for .maxnctapersm directive. Removing benchmarks with known-issues from regression list for now.
[git-p4: depot-paths = "//depot/gpgpu_sim_research/fermi/distribution/": change = 14572]
Diffstat (limited to 'src')
| -rw-r--r-- | src/cuda-sim/instructions.cc | 16 | ||||
| -rw-r--r-- | src/cuda-sim/ptx.y | 1 |
2 files changed, 16 insertions, 1 deletions
diff --git a/src/cuda-sim/instructions.cc b/src/cuda-sim/instructions.cc index e7bb50a..99d7399 100644 --- a/src/cuda-sim/instructions.cc +++ b/src/cuda-sim/instructions.cc @@ -1452,6 +1452,20 @@ ptx_reg_t sext( ptx_reg_t x, unsigned from_width, unsigned to_width, int to_sign return x; } +// sign extend depending on the destination register size - hack to get SobelFilter working in CUDA 4.2 +ptx_reg_t sexd( ptx_reg_t x, unsigned from_width, unsigned to_width, int to_sign, int rounding_mode, int saturation_mode ) +{ + x=chop(x,0,from_width,0,rounding_mode,saturation_mode); + switch ( to_width ) { + case 8: if ( x.get_bit(7) ) x.mask_or(0xFFFFFFFF,0xFFFFFF00);break; + case 16:if ( x.get_bit(15) ) x.mask_or(0xFFFFFFFF,0xFFFF0000);break; + case 32: if ( x.get_bit(31) ) x.mask_or(0xFFFFFFFF,0x00000000);break; + case 64: break; + default: assert(0); + } + return x; +} + ptx_reg_t zext( ptx_reg_t x, unsigned from_width, unsigned to_width, int to_sign, int rounding_mode, int saturation_mode ) { return chop(x,0,from_width,0,rounding_mode,saturation_mode); @@ -1748,7 +1762,7 @@ ptx_reg_t (*g_cvt_fn[11][11])( ptx_reg_t x, unsigned from_width, unsigned to_wid int rounding_mode, int saturation_mode ) = { { NULL, sext, sext, sext, NULL, sext, sext, sext, s2f, s2f, s2f}, { chop, NULL, sext, sext, chop, NULL, sext, sext, s2f, s2f, s2f}, - { chop, chop, NULL, sext, chop, chop, NULL, sext, s2f, s2f, s2f}, + { chop, sexd, NULL, sext, chop, chop, NULL, sext, s2f, s2f, s2f}, { chop, chop, chop, NULL, chop, chop, chop, NULL, s2f, s2f, s2f}, { NULL, zext, zext, zext, NULL, zext, zext, zext, u2f, u2f, u2f}, { chop, NULL, zext, zext, chop, NULL, zext, zext, u2f, u2f, u2f}, diff --git a/src/cuda-sim/ptx.y b/src/cuda-sim/ptx.y index c4d8b06..15662aa 100644 --- a/src/cuda-sim/ptx.y +++ b/src/cuda-sim/ptx.y @@ -218,6 +218,7 @@ block_spec: MAXNTID_DIRECTIVE INT_OPERAND COMMA INT_OPERAND COMMA INT_OPERAND {f func_header_info_int(",", $4); func_header_info_int(",", $6); } | MINNCTAPERSM_DIRECTIVE INT_OPERAND { func_header_info_int(".minnctapersm", $2); printf("GPGPU-Sim: Warning: .minnctapersm ignored. \n"); } + | MAXNCTAPERSM_DIRECTIVE INT_OPERAND { func_header_info_int(".maxnctapersm", $2); printf("GPGPU-Sim: Warning: .maxnctapersm ignored. \n"); } ; block_spec_list: block_spec |